Skip to content
Agents tracked: 492 Releases (7d): 545 Agent status: 1 with issuesSince last limit reset: Claude 9d · Codex 7d For AI agents: Markdown Updated Nov 20, 2026

AI agent releases: Sep 28 – Oct 4, 2026

429 stable releases from 121 tracked agents and frameworks in the week of Sep 28, 2026, newest first, with the headline changes from each set of release notes. Our lists keep each project's latest releases, so a very active project's older releases can be missing.

Oct 4, 2026

  • Agent Mail 连接器(授权、邮件工具、新邮件任务)
  • LDAP 目录登录
  • 专家默认对话模式;输入栏「+」菜单;未开审批时隐藏批准入口
  • Hooks save and return. The observation, file-edit, Stop and SessionEnd hooks write each event to a file under ~/.claude-mem/state/hook-spool/ and exit, without waiting on the worker or starting it.
  • SessionStart context is ready before you start. The worker prepares SessionStart context in ~/.claude-mem/state/context-cache/, refreshing it about 2 seconds after anything that could change it…
  • Observer prompts that providers can cache. A provider's prompt cache (OpenRouter's, or the cmem.ai gateway's) can only reuse the part of a prompt that starts exactly as before.
Kortix
  • Fork a conversation. A session's conversation offers a Fork action, so you can branch from any point.
  • Cancel a subscription from the billing pane.
  • One permission per topic. The broad project.customize.* permissions are replaced by project.settings.write, project.sandbox.write, and project.model.read/.write.
  • Terminals launched through the shell tool now stop when fx exits instead of staying alive for the next resume.
  • New libfx checkpoints cannot be restored by older versions. Existing checkpoints still load.
  • Invalid Base64 image input now throws directly from prompt() instead of failing the turn later.
Experiential
  • Cache-placed sessions stay on their rung: when a host's cache placement names a rung that authors a concurrency bound with saturation=refuse, a per-worker shed there…
  • Workflows: write rules under workflows.rules in config.json that act on pull request events in the repositories you name.
  • Keep mergeable: toggle Keep mergeable in Reviews > Mine to have a sandboxed agent fix merge conflicts or failing checks on a local copy of that PR.
  • Merge when ready: toggle Merge when ready on one of your pull requests in Reviews > Mine to add it to Glimmervoid's own merge queue.
Graphify
  • Feature: four more language extractors gained type coverage from @rajatnagda45 — Zig tagged-union variants (union(enum)) emit a node per variant with a case_of edge, C++ union specifiers are…
  • Feature: three Scala type-reference fixes shipped together earlier in the day are followed here
  • Fix: Elixir import/use call scoping is now per-module rather than per-file, so an unqualified call in one module no longer resolves against another module's imports in the same file…
  • delegation: send inbox replies on their own under Handle it for me (23af959)
  • orchestrator: let the Executive take the lead behind one approval gate (f341051)
  • ui: harden the UI for phones, readability and plain wording (d46eee5)
  • New providers and balances: WorkBuddy (Tencent) joins as provider #90 with its monthly credits allowance; Claude shows promotional cloud-session credits separately from prepaid credits.
  • Antigravity works again on 2.19.1: Login with Google no longer fails with invalid_client (discovery now pairs each OAuth client id with its own secret), and additional Gemini profile homes can be…
  • Sign-in and cookies: explicit refreshes can prompt for Keychain access again for every plugin cookie provider, the CLI explains rejected sessions instead of "missing cookie", SweetCookieKit 0.5.5…
nanoMuse
  • Holds: the hands are handed over, and wait. A hold belongs to one chat and one kind of hands — the browser, this computer, the phone.
  • Approvals where you are. The desktop's live stage carries Allow once / Always in <app> / Deny (always = a per-application grant, revocable on the agent page).
  • A browser you can drive (web). The viewer takes over, clicks, types, scrolls, opens a URL, goes back, says paused, waiting for you while the agent waits, and Done hands it back.
  • Fixed native OpenAI context compaction for sessions containing many screenshots, preventing image-size estimates from incorrectly causing compaction requests to be rejected.
  • Fixed compatibility with Command Code DeepSeek and other DeepSeek-family models by preserving the reasoning context required for warm OpenAI Responses sessions and correctly handling incomplete…
  • Fixed waiting for subagent follow-up messages: responses now appear as background jobs that can be waited on or canceled, and are delivered only once.
Billion Context
  • feat: publish merged-to-master builds as billion-context@master
  • fix: stamp host-native ACP tool calls as native callers (Pi subagent cross-session misrouting)
  • docs: pi 原生模式传输覆盖说明与 WS 绕行配置
  • 侧栏大改版:新增信息(ⓘ)、轨迹(∿)、工作目录(⌗)三个面板,面板标签支持点击与悬停、⤢ 一键全屏,侧栏文字可直接框选复制。
  • 宠物与启动页吉祥物:宠物沉底常驻、可拖动互动、通知改为气泡、新增鲸娘皮肤;启动页也有吉祥物坐镇。
  • 主题实时预览:主题选择器旁实时渲染代码与工具卡预览,改主题不必再来回试探。
  • fix(cdp): flush Runtime enable replay before its response #917
  • fix(devtools): misc followup for cdp response ordering #1017
  • fix(cdp): misc followup for cdp refactor #1047
  • 0feded28b8aa13a03ff722f70bf7890415b1d1e7 fix: handle missing organization and user in remove-user-from-group API
  • 29c245172e8b4556bbae37473f89e9bf9a6a9a4b fix: mark JWT-Custom refresh tokens with tokenType refresh-token
  • 0cef429dabdcd9a371d09d15300f10ddc9726a8d fix: redirect to the requested page after login
  • Native chat (experimental): The "Stop" button ends Claude's process, including background commands and subagents. If Codex refuses or never answers a stop request, Orca ends it too.
  • Chat polish: /clear switches chats immediately; the first message starts the new agent. Codex connection retries update one warning instead of stacking errors.
  • Agent status: Esc cancellation settles Codex panes promptly, while Ctrl+C used to copy or leave /side no longer looks like an interruption. OpenCode shows failed and stopped turns correctly.
Caveman
  • maintenance/2026-09-20: /caveman status, defaultMode: "manual".
  • Contributor PRs merged: #1078 (C++ parse gate), #1076 (GDScript elision), #1156 (syslog severities), #1108 (delete dead proxy/routing), #1084 (Grok Build install, closes #547), #1062…
  • Daily-triage remnants cherry-picked: Python encoding fixes, merge scope check, the conformance pin-bump chain.
  • feat(library): Slack AI Agent Security: Permissions, Data, and Audit Guide
  • feat(library): Best HubSpot Alternatives for AI Marketing Automation in 2026
  • feat(library): How do you build an AI voice agent with Twilio and Sim?
  • Session replay. Replay recorded sessions step by step and discuss the recorded steps with your agent.
  • Connector growth. New Cellosaurus and Monarch connectors arrive, and GEO matrix discovery joins the omics connector.
  • Uploaded PDF structure extraction. Structure extraction now works on uploaded PDFs, not only generated ones.
  • fix(plan): run a lone update_plan instead of refusing it (6682d92bf) — Vincent Hus
  • perf(onboarding): isolate repair Codex home, warm the first turn, apply /effort (96836016b) — Vincent Hus
  • test(slack): clear JWT_TOKEN in the Slack app-connect demo test (bf72f9f99) — Vincent Hus
  • Sampling by thinking level — samplingParamsByThinkingLevel in models.json sets sampling parameters such as temperature and top_p for each thinking level on OpenAI-compatible APIs.
  • Added samplingParamsByThinkingLevel to models.json for per-thinking-level sampling parameter overrides on OpenAI-compatible APIs. See Configure sampling by thinking level

Oct 3, 2026

  • Fixed a deny or ask rule on a nested part of a compound shell command not holding over a user-installed mod's approval on managed machines
  • Fixed the terminal freezing on short code blocks with many unclosed <script> tags or deeply nested ${ substitutions
  • Fixed Read deny rules not applying to files @-mentioned, changed, or selected in the IDE through a symlink
  • chat: fold messages sent mid-turn into the running answer (bba990f)
  • delegation: check Act as me drafts against what the writer already told those people (0543e92)
  • memory: bring Always in the loop notes into the owner's briefs and remind when due (a5b399d)
Oh My Pi
  • When a DeepSeek turn ends with raw <|DSML|…> tool-call text that could not be parsed into a real call, and no tool call was made, the broken markup is now removed from the message before it is…
  • Fixed Antigravity chat and image requests sending an outdated client version when the model list came from cache, which could make newer models such as Claude Opus 5.5 unavailable.
  • When a DeepSeek model writes a broken DSML tool call (for example with the opening <|DSML|tool_calls> and <|DSML|invoke> tags missing), its closing tags are now kept in the streamed text instead…
  • Profile: basics, skills and languages, experience, education and references, with a strength meter. Start it from an existing resume (PDF, Word or Markdown).
  • Base resume: six themes and a theme builder, previewed on the pages they print on and rendered to PDF on the server.
  • Find jobs: search postings through JSearch, Adzuna or Germany's Bundesagentur für Arbeit, save or dismiss them, and start an application from one.
  • HTML report. A saved report also includes complete_report.html, the report as one page with its sections listed beside the text, that reads well in a browser, on a phone and in print.
  • A provider per model tier. The quick and deep models can run on different providers, for example the managers on Claude while the analysts run on OpenAI: set deep_think_provider or…
  • Past decisions settle alongside the analysts. While a run's analysts work, every logged decision whose holding period has passed is settled, for any ticker, not only the one being analyzed.
  • Your computer, on the web. Each physical computer appears once, even after you re-pair it.
  • Edit a queued prompt in place. The prompt queue sits in its own card above the composer.
  • Review Center has its own page. You approve requests in a modal on the approve page.
Billion Context
  • release v0.1.182
  • docs(release-notes): add v0.1.182 entry
  • ci: add dedicated bugfix release channel (carve-out publisher,
  • commit: 88e224475a6183f7b31218a7499be956a4bf2667
  • source tag: beta2026-10-03_23-39-47
  • regions: cn global
  • Nix flake — nix run github:earendil-works/pi/stable runs the latest release, and nix profile add github:earendil-works/pi/stable installs it. See Install pi.
  • Project overrides for MCP servers — .pi/mcp.json and /mcp can enable, disable, or change the exposure of a user-level server for one project. See Configure servers.
  • MCP Client ID Metadata Documents — oauth.clientRegistration: "cimd" lets authorization servers allow pi by its document URL instead of dynamic registration. See Authenticate with OAuth.
  • Agent chat: attach files with the new Add files button or by dropping them onto the message box, on the home screen as well as in the chat. The model picker is more compact.
  • Upgrade prompt: the editor window now comes forward when an upgrade prompt opens, so it can't stay hidden while the app runs in the background.
  • Alt-drag: holding Alt while dragging duplicates the element, and the copy now moves reliably with the pointer.
  • fix(vps): accept operator resource and restart settings
  • Connect your phone: one menu entry, to this computer or your Cloud
  • fix(server): make late VM completion overlap deterministic
Experiential
  • Gateway Responses admits include reasoning.encrypted_content on a route whose every rung carries one gateway reasoning-carrier scheme, including rungs declaring reasoning_content_native…
nanoMuse
  • No token in any URL. The socket is opened bare and the token is its first frame ({"kind": "auth", "token"}, within ten seconds, or 4401).
  • The device being controlled agrees first. Another device of your account that wants to run, read or write something on this one over the hub…
  • Approvals, the same on every side. device_files and device_get ask now (they were safe).
  • Lighter cost tracking: unchanged Codex refreshes reuse the decoded local history instead of decoding the whole cache, empty session histories stop being rewritten on every pass, Claude keeps…
  • Codex weekly quota recovers after a reset: when the previous snapshot had no credit inventory, the fresh quota, reset credits, and update time are published instead of "1% left" lingering for hours.
  • Clearer cookie sign-in: every browser-cookie provider now names the browsers automatic import supports and offers Manual for the rest, instead of saying "Chrome" regardless.
  • No more lost first answers. When ChatGPT is slow, a new chat's first answer used to show up above your question, without Copy or Export, and the app reported "Stopped waiting for delivery…
  • Sending into existing chats works again. ChatGPT changed its page, and sends failed with "Requested model or reasoning could not be confirmed". Fixed.
  • Watch the first answer take shape. While ChatGPT works on a new chat's first answer, the app shows what the model writes between its steps, instead of nothing until the end.
  • fix(mcp): derive model output schemas from serialization
  • fix(tenki): upgrade sdk and preserve timeout failures
  • chore(deps): bump gitpython from 3.1.59 to 3.1.62
  • Pi 1.0. Feynman now runs on Pi 1.0.0, with pi-subagents 0.75.0 (background subagents start on Pi 1.0) and pi-btw 0.7.1 (allows Pi 1.x).
  • Two MCP tools. work_state_write appends one entry to a named list (fields capped at 2,000 characters as JSON) and replies with what is still open in that list.
  • Lists, items and state. An entry with a task field is a to-do item with a status of todo, doing, done or dropped; any other field is state on the list.
  • SessionStart leads with it. Context opens with the rule and every open item, each with how long ago it last changed, capped at 3,000 characters (806 when nothing is open).
Sim
  • feat(forks): compare last synced source deployments
  • feat(youcom): add You.com integration
  • feat(search): browse Lucid folders and Notion pages
  • Added ←/→ shortcuts to /config: on a highlighted setting they switch to the previous or next value and save it right away, without opening the dropdown, and the footer shows a ←/→ Change hint
  • Improved responsiveness in long conversations in no flicker (altscreen) mode: resizing the terminal no longer freezes or lags the CLI, the full-screen view redraws the messages on screen first…
  • Improved mouse text selection in the altscreen view and in the artifact viewer: dragging past the top or bottom now auto-scrolls, the selection stays on its text while the agent streams or you…
  • Release notes — formatted for people, with expandable sections.
  • Changelog — plain Markdown for AI agents and tools.
  • npm package
  • Handle draft-7 list-form items in JsonSchemaTransformer, TestModel and Mistral streamed output, and make TestModel accept boolean subschemas and cap prefixItems at maxItems
  • Fix ImageGeneration under FallbackModel and dataclasses.replace, and compare Embedder by identity
  • Refuse a second durable execution engine on an agent before either binds
  • b94dce99fa1d9fd21de82e1c93b50640f9d3968f feat: open account menu on hover
  • bd130188f034625f76d3b1e51f15340d8e33ca81 feat: return stable error code in login failure response
  • 8f7efcf0d89ea98dc21db1b15c21b5bdcd8384c9 fix: check terms of use by default on login page
Omnara
  • ghcr.io/omnara-ai/omnara-web@sha256:7e8d1bbbd11b16e0533bc1dc24fbc91f43a652549d6ab67f27a84ceec954f733
  • ghcr.io/omnara-ai/omnara-api@sha256:e441e5ca30f73dedffd06bf36f3a650000b298381166cc46dd564a98885528b7
  • ghcr.io/omnara-ai/omnara-worker@sha256:1f088f1c4a0ef75e4bc63e0f545ac766d3a0ac0108d595ae5419b19f7aca8336
  • Select GitHub connections across shell sessions and cloud prompts (aabde6f29) — Vincent Hus
  • Record Skill A business value after successful display (42a17c4ea) — Vincent Hus
  • feat(skills): live skills releases pulled every five minutes (956e8b27f) — Vincent Hus

Oct 2, 2026

  • bb desktop 0.45.0
Superset
  • fix(desktop): make host connection banner visible and fix cloud-wake copy
  • feat(db): task_imports table
  • fix(desktop): reorder cloud workspace context menu sections
  • feat(react-core): make intelligence indicator auto-mount configurable (a161b38)
  • feat(learning): capture raw browser activity as AG-UI events (1840e7c)
  • feat(mcp-apps): consolidate the Vue and Angular hosts onto the shared package (06f61ae)
OpenExecutive
  • auth: add SSO sign-in through any OpenID Connect provider (d338f1f)
  • delegation: let Act as me use an Outlook mailbox (6b7957e)
  • integrations: add a Microsoft 365 (Outlook) mail and calendar backend (5ec59a4)
  • feat(terminal): warn when a typed Codex joins Codex's shared server (STA-9051)
  • feat(terminal): point an old terminal's Codex shared-server banner at a new terminal (STA-9051)
  • fix(terminal): give plain fish tabs Orca's codex function without changing fish's startup
Cline
  • Existing users get a one-time What's new dialog introducing Connectors, with connector logos, a clickable example prompt, and an Open Connectors button.
  • Added $.ui.selection() for mods: returns the text you last selected in fullscreen mode and, when the selection lies within one transcript row, that row
  • Added a built-in gh api to cloud sessions whose image has no GitHub CLI, and fixed the built-in sending control characters from file names, jq filters or GitHub errors to the terminal
  • Added recovery for a prompt cleared with Ctrl+C: pressing Up on the empty prompt brings the draft back, including pasted text and images
OpenMausBot
  • Companion: recover a reclaimed tunnel automatically, even while "ready"
  • ci: keep main to one running CI, and give releases their own CI lane
  • Integrate reviewed tools, skills and workspace fixes (Oct 2 batch 2)
  • #966 0f2060f Thanks @waseem-k-08! - Add a GitHub Copilot CLI integration: openwiki integrations install copilot.
  • #975 c5d7a6d Thanks @colifran! - feat: add entra auth configuration and token provider
  • #972 1d09d95 Thanks @eugeneliu-86! - feat: group a repository run's planner and page workers into one LangSmith thread, named "planning agent" and "worker agent: <page>"
  • Resizable Picture-in-Picture window for MCP Apps
  • Fetch model metadata from models.dev live with bundled fallback
  • Quoting action to append selected conversation text to composer
nanoMuse
  • A first run like Muse's. Full-window; welcome → sign in with a phone number or an e-mail → the code (or a password, /v1/auth/login) → the permissions carousel with ‹ › → ready.
  • The window, screen for screen. Rail (Chats · Search · Devices · Schedules when installed · the hamburger), the chats column (Search, Main chat, Side chats with +.
  • Muse's permission card. The harness's approval card restyled — shield, headline, Allow once in blue, Reject — and every answer recorded in the profile panel's Approvals tab.
  • #2451 df9c0ef Thanks @aron-cf! - Update pi-durable and pi-ai to ^1.0.0. See Pi harness docs for details.
  • Two new providers: Muse (muse.ai) and LithosAI read your own subscription usage and balance with your browser session, bringing CodexBar to 89 providers.
  • Safer by default: CodexBar never runs binaries from the current directory or relative PATH entries, and Codex discovery refuses unsafe Node environments and PATH entries before launching anything.
  • Much lighter cost tracking: a long series of CPU and allocation cuts for Claude, Vertex, Codex, and local agent logs, plus cached Gatekeeper verdicts for unchanged CLIs. Thanks @dustball!
  • Added prebuilt native Android binaries of the CLI that can run directly in Termux, without requiring a proot-distro environment.
  • Improved responsiveness while typing, scrolling, and streaming agent output by sharply reducing the memory the CLI allocates on every update and redraw
  • Improved -p "/usage" to show quota reset times in your local time zone when printing to a terminal; piped and --json output still use UTC timestamps, so existing scripts are unaffected
Openbot
  • Add a ChatGPT dots comparison and show where each product works
  • Let local scripts run an agent's routine with a payload
  • Draw the content artwork as WebP
Sim
  • improvement(audits): enforce console, helper, render-path, persist, and deployment-flag rules
  • fix(deps): bump mammoth to 1.12.3
  • fix(logging): name the driver cause and redact bound params in logged errors
  • Feature: four more language extractors gained structural depth. Ruby paren-less self-sends (do_thing with no receiver or parens) are now captured as calls, with a…
  • Fix: a Rust type used before it is declared in the same file now resolves to its local declaration instead of fabricating an external stub.
  • Fix: Rust prelude types (Option, Result, Vec, String, Box, Rc, Arc, HashMap, and friends) no longer accumulate spurious in-degree and surface as god nodes that distort community detection and…
Oh My Pi
  • Added createAuthGatewayRouter, the auth-gateway's routes without the HTTP listener, and serveAuthGatewayStdio, which serves them as JSON lines…
  • Added omp auth-gateway stdio: a long-lived inference server for other programs speaking JSON lines on stdin/stdout ({"id", "path": "/v1/chat/completions", "body"} in, {"id", "status", "body"} out)…
  • Reduced CPU use while streaming with several agents active: extension message_update handlers are delivered through a lighter queue, and RPC no longer processes subagent events unless a client…
  • Gitea webhooks now require a secret: with gitea.webhook_secret unset, the Gitea app rejects every webhook with HTTP 403 instead of accepting it unsigned.
  • The GitHub Action fails on tool errors: a tool failure that used to be logged and swallowed now fails the Action. Set github_action_config.fail_on_tool_errors = false to keep the old green behaviour.
  • Comment commands need a leading / on GitLab, Bitbucket and Bitbucket Server: a comment only runs a tool when it starts with a slash command, so plain comments that mention one no longer trigger…
OpenAI Agents SDK
  • restore release tests and refresh required readiness check (730deb4)
  • The clean candidate contains current main a575a6e637feb9aea1b591237b007dd4991ddfba plus only six release-owned metadata files.
  • Relative to v0.23.0, SDK runtime source changes only its version string. API snapshot content, dependencies, Python support and persisted schemas are unchanged.
  • arXiv and PubMed rate limits. arXiv asks for one request every three seconds, and parallel lookups got 429; arXiv requests now run at that pace and a 429 is retried once.
  • Running several subagents at once from any workflow no longer fails with "workflow: true requires exactly one js workflow fenced block": Feynman now always writes the workflow script to a file and…
  • Pi 1.0 is not in this release: pi-subagents 0.74.0 cannot start background subagents on it.
  • agent-framework-foundry-hosting: Host native workflows over Foundry Responses and Foundry Invocations
  • agent-framework-duckdb, agent-framework-sql-server, agent-framework-typesafe: Add DuckDB and SQL Server native vector-store connectors and the TypeSafe AI connector
  • agent-framework, agent-framework-core: Add response-stream gates and buffering, session-scoped file-access isolation, an AgentExecutor checkpoint-state TypedDict, and opt-in…
OpenViking
  • 检索与排序:每条查询改为一次全局向量召回,再按模式统一 rerank,移除目录递归与父级分数传播,评分阈值在 rerank 之后应用(#5450);移除热度加权及 retrieval.hotness_alpha(#5454);find / search 新增事件时间衰减排序参数 events_time_decay_protection(#5214);search 新增…
  • 文件系统与知识整理:ls / tree 返回 has_more 分页状态(#5330);tree 支持 directories_only 目录过滤和 L0/L1 展示(#5334);ls 支持 include_abstract / include_overview(#5434);ov compile --skill memory…
  • MCP、权限与 Studio:MCP 工具声明行为注解(#5075),整次调用失败时返回 isError: true(#5078);新增 list_users、list_groups、get_acl、set_acl 工具,write / add_resource 支持 acl 参数(#5466);账号变更缺少注册表基线时拒绝写入,避免覆盖已有账号(#5483);ovcli.conf 接受…
OpenClaw
  • GPT-6.1 Sol support: add the current Sol model across OpenAI routing, discovery, reasoning, harness, and Reef guard-model boundaries.
  • Safer updates and recovery: preserve plugin settings, prevent duplicate Gateways, handle pnpm package updates without terminal failures, and retain plugin inventory through migration.
  • Reliable agent completion: preserve complete delegated and CLI answers, release suspended child capacity, recover full cron output, and prevent failed requests from consuming steered questions.
  • Built with WXT: the extension is now a single package built with WXT instead of a multi-package
  • New AI engine: LLM calls now run on the Vercel AI SDK instead of LangChain. Structured output is more
  • Reasoning effort: models that support it now use a reasoning effort setting instead of temperature
  • [OPIK-7965] [FE] feat: render compare prompts as messages and store prompt templates
  • [NA] [SDK] fix: count Gemini tool_use_prompt_tokens in prompt_tokens
  • [NA] [SDK] fix: record Azure OpenAI haystack components as provider=azure
  • Live progress & cancellation for page sync: Knowledge.stream_sync_pages() / astream_sync_pages() yield typed PageSyncProgress snapshots plus one terminal SyncReport, and sync_pages /…
  • Page sync reports name failed pages: SyncReport.failed_paths lists the site paths of pages that failed to publish/delete (capped at the first 20, like errors; failed keeps the full count.
  • Retry transient page-fetch failures with backoff: on large corpora (e.g. docs.agno.com, 3,913 pages) 1–2 random pages failed per run with Connection reset by peer.
  • Added a confirmation before a pairing link connects to a new host or to a saved host whose key or relay changed
  • fix(board): converge aged reuse/recovery entries off the board render
  • fix(task): keep checkpoint snapshot ids unique across restarts
  • fix(wake): defer periodic wakes while v2 children make progress
  • feat(web): show error id in tRPC error toast and include it in issue reports
  • feat(evals): add facet type, built-in protection, and rule idle time
  • feat(evals): show seven-day execution health in evaluator status
Experiential
  • Local capacity refusals can retry the same idempotency key only after a durable certificate proves that no paid effects occurred.
  • Nonempty compatible reasoning frames update internal generation progress without exposing private reasoning. Overall request deadlines remain enforced.
  • The grounded-world prompt interns repeated task and context content without summarizing examples. Packing preserves required content, output allowances, and included-example provenance.
  • fix: fork-adoption 收养管线补齐 responses/google 两线(resume 继承 refs/blocks)
  • fix: keep estimate-grade values out of context sizing denominators
  • fix(loop): Responses degenerate retry honors custom_tool_call and keeps every attempt's usage
Huobao Drama
  • 统一火宝品牌图标:网页 Logo、浏览器图标、macOS 应用图标和 Windows 应用图标。
  • 提供 macOS Apple Silicon、macOS Intel 和 Windows x64 安装包。 macOS Apple Silicon 安装包内服务与前端启动检查通过;其他平台已完成打包和内容校验,未进行对应系统运行测试。
  • core: Keep the task runner running on an unhandled promise rejection (d719c87)
  • Answers are recorded again for messages with the Core mention. Since 2.1.24, CoS ends its messages with an "@Chat On Steroids Core" mention.
  • Chats continued in a Project are followed correctly. After Compact & resume in a Project, ChatGPT keeps the old chat hidden in the same tab.
  • An automatic Continue no longer gets stuck in the message box. After a reload, CoS's own "continue" text could stay unsent in the message box and block every further repair, until you pressed Send…
  • WebView perception works the same everywhere — mobile WebView runs now use the same Appium bindings as the rest of the mobile stack, so what the agent sees matches what's on screen.
  • Exports pick the right platform automatically — the export platform is derived from the session target you locked in, so a mobile run no longer produces desktop code.
  • No more spurious export errors on mobile — code export validation is skipped for mobile runs instead of failing them.
  • GHSA-6fqq-452j-qhrp (high): a streamed request through ConcurrencyLimitedModel or limit_model_concurrency could keep its concurrency slot when the slot was released on a different task than the…
  • Add oneOf schema support to TestModel's generated data
  • Deprecate pydantic_ai_harness.step_persistence.naming
  • The legacy Composio broker is retired. The API-key catalog method, broker clients, session creation, account synchronization, child lifecycle and toolkit routes, and the Services tab are all removed.
  • Execution harnesses now default to full auto. Native runs default Claude/ACPX to approve-all, OpenCode to allow, and Codex to approval-and-sandbox bypass, and the legacy adapters follow the same…
  • Agents get faces — Static persona images replace agent icons across lists, the sidebar, org charts, tasks, comments, selectors, activity, and dashboards, with one animated pointer-following…
  • Newest models: Claude Sonnet 5.5 with adaptive thinking, GPT-6.1 Sol, an Astra Ultrafast speed switch, and automatic fallback when a gateway model refuses or is unavailable.
  • Approvals got stricter and clearer: the approval card is now the only way to approve credential use, strict-posture approvals show the gated call's arguments, and single-use credential grants are…
  • Security screening simplifies to three modes: off, observe, and enforce. Memory adds provenance-aware disclosure and audience isolation.
  • Fix/gateway unanswered grace (e53433bf7) — Vincent Hus
  • Stamp Slack member and workspace on gateway turns (50931e9f2) — Vincent Hus
  • Let the scheduler own CI repair ticks instead of a forced /cron run (6092fcf5f) — Vincent Hus
  • Execution cancellation: Pass an abort_signal to Runner, Workflow, and nodes to stop a run gracefully; /run_sse now cancels the run when the client disconnects. See the unit guide for more details.
  • Tool confirmation in workflows: Tool nodes now pause for user approval via RequestInput, the same way an LlmAgent does, instead of passing an error downstream. (ce132b9)
  • ModelConsultTool: Let an agent consult another model mid-task, capped by per-turn and per-session budgets. See the unit guide for more details. (84cc99a)

Oct 1, 2026

Omnara
  • feat(machines): add Tenki sandbox provider
  • feat: add sandbox provider docs pages
  • docs: describe the default machine environment for Omnara-managed pools
  • pi/core no longer expands the project name: its README and the bundled
  • Downstream packagers need no changes. The package is still ecc-pi-core, the
  • After npm publish, the release workflow polls the registry for up to five
  • datagen: send each recorded application to its own project (d99a4db)
  • tracing: add the DECISION span kind across the platform (43d1a54)
  • honor external resources flag for WASM and UI (7a71a21)
  • feat(smallestai): add TTS continuations protocol support
  • baseten: inline mid-conversation instructions for Qwen/Qwen3.8-27B
  • fix: drain inference STT transcripts before closing
  • Amazon Q for VS Code 2.8.0
Deep Agents
  • Emit a cache_expiring hook before the prompt cache expires.
  • Keep the transcript at its current position when opening pickers.
  • Share cache-expiring notifications with the TUI.
  • fix(collector): read spreadsheet dates as dates, not serial numbers
  • Drop temp param for Anthropic models
  • Allow Generic OpenAI max tokens of 0 to omit max_tokens from request
Feynman
  • Launch warnings and the first-run telemetry notice show inside the session. Feynman printed them to the terminal just before starting Pi, where Pi 1.0's fullscreen interface…
  • feynman doctor reports Pi's new quietStartup: "header" setting as "header only" instead of "disabled".
  • Give a tokenless Copilot Accept an outcome the client can read back (SKY-17870) (717519a8f)
  • security: patch 15 Dependabot npm alerts in skyvern-frontend and the TypeScript SDK client
  • Fix the TS SDK npm release: stop .gitignore from swallowing the artifacts resource
  • Fixed auth vault login to pick usable controls when hidden duplicates exist, and to bind credential entry to the selected field so replaced fields or redirected focus are rejected before submitting
  • Fixed chat mode to allow webmcp, skills, and other top-level browser commands, so the model uses page WebMCP tools instead of falling back to eval
  • Fixed Kernel profiles so KERNEL_PROFILE_NAME sessions send the profile as an object instead of failing with a 400, and added KERNEL_PROFILE_SAVE_CHANGES to opt into saving changes back
Kortix
  • Microsoft Teams is on for every project. There is no feature flag to turn on: connect Teams from the project's Connectors → Channels page.
  • What Kortix says to one person in Teams stays with that person. In a channel or group chat, only that person sees
  • the prompt to connect an account or request access, with the connect button right in the thread;
Chat ON Steroids
  • Compact & resume works again in Projects. For chats inside a ChatGPT Project, every Compact & resume failed with "ChatGPT could not open the source Project".
  • Workers and Loop get their tools on every account. On some accounts (seen on Plus and Business in ChatGPT's Chat mode), ChatGPT only gives a chat the Chat On Steroids tools when the message…
  • The Plugins connector refreshes again. With larger plugins installed, such as Unity, the automatic refresh of the Plugins connector failed every time, so ChatGPT kept an old tool list after an…
Experiential
  • Catalog-backed evaluation spending limits are now opt-in. New evaluations display their estimated cost with the spending limit off.
  • The SDK accepts None for the aggregate evaluation or request-ledger limit. Per-request token and retry-inclusive cost reservations remain finite, and completed charges, unknown-dispatch holds, and…
  • EvaluationServices preserves its prior omitted-argument behavior using the execution budget; explicitly passing spending_limit_usd=None removes only the aggregate cap.
  • @luyanfeng, for reporting the OpenCode 2 loading failure in #298.
  • 本地语音听写:输入框旁新增麦克风按钮,点击录音、再点结束。音频重采样为 16 kHz 单声道 PCM16 WAV 后交给本机识别进程转写,文字进入输入框草稿、不会自动发送;草稿已被改动或正在输入中文时,结果会挂在「插入」按钮后面等你决定。「设置 → 语音输入」可开关、选语言和麦克风,并试听查看波形。识别引擎和模型运行时下载到本地(校验…
  • 工作区文档预览:Agent 写出的 PDF、Word、Excel 直接在工作区侧边打开,不再跳到别的应用。支持缩放与适应窗口、按文件记住滚动位置、缩放和当前工作表,文件被改写后自动刷新;聊天里 Agent 提到的本地图片(工作目录之外、~、C:\、file:// 路径)直接显示,Read 等工具返回的图片以缩略图出现在调用下方。工作区之外的文档仍交给系统程序打开。@olo12119(#1397)
  • 精选技能目录:技能市场默认打开内置精选目录(398 个 ClawHub / SkillHub 技能,13 个分类),不再每次实时请求两个注册表;「搜索全部市场」保留为显式范围,结果会标注「未精选」。详情页补齐统计、安全报告、更新日志、能力与触发条件;未审计或被标记有风险的技能需要确认后才能安装。ClawHub 的技能固定按卡片上的作者安装,避免同名技能被别人顶替。
  • Fullscreen by default — The TUI now runs fullscreen. Set tuiMode to "regular" to keep the terminal's normal scrollback. See Terminal and display.
  • Leaner codemode — About 40% fewer prompt tokens, and errors that tell the model how to recover. See Codemode.
  • Image generation in codemode — Scripts call models.generateImages() with the session's credentials. See Generate images and Use image models.
nanoMuse
  • The phone in the browser takes the web version's place. nanomuse.cn/web leads to the showcase: a simulated phone with a private Muse, started after a sign-in to nanoMuse Cloud…
  • DeepSeek V4 Pro answers in the showcase. MAIN_MODEL defaults to deepseek-v4-pro on 阿里云百炼.
  • A demo is called a demo. The card beside the phone — the phone and its apps are MobileGym's re-creations, the nanoMuse in it the lite web app, a long way from the Android app — with Get the…
  • feat(bidi): graduate bidirectional streaming API (8dbc128ea)
  • feat(vended-tools): port a2a_client tool to TypeScript (ff114a9bd)
  • feat(bidi): display tool calls in console I/O (5a940fbae)
  • Added Claude Mods: plugins may now modify deeper behavior
  • Added You should know, a built-in mod where a side agent watches your back and flags things you or Claude might miss.
  • Added an n:<text> filter to the agents view that matches session names and tasks; a filter now shows matches in collapsed sections and Enter opens the first match
  • feat(library): What are the best marketing automation platforms for AI workflows in 2026?
  • feat(library): AI Agent Marketplace vs Building From Scratch: Where Sim Fits
  • improvement(search): run Search retirement as an operator command instead of a deploy step
  • Web search now uses your model provider's own server-side search (Anthropic and OpenAI) by default instead of scraping from your machine, shown as regular web search rows live and on resume…
  • Set websearch.prefer_native = false or JCODE_WEBSEARCH_PREFER_NATIVE=0 to keep using local search engines
  • Provider-side search follows each session's allowed and disabled tool policy
  • 💻 Windows:双击下载的安装包,按提示完成安装。
  • 🍎 Mac:打开下载的安装包,将 toonflow.app 拖入“应用程序”,再从“应用程序”打开。
  • feat(review): default agent-triggered review off
  • chore(deps): upgrade pi 0.87.1 → 0.99.1
  • feat(web): render Skills state with switches
  • The PageIndex SDK, local or cloud — vectorless, reasoning-based RAG, end to end.
  • Much faster indexing — the PageIndex Flash engine gets the tree from layout stats: no LLM involved for the structure generation itself, LLMs only write the node summaries, and tree expansion…
  • Flash engine: the local default (mode="standard" keeps the classic LLM pipeline).
Oh My Pi
  • Fixed rejected HTTP 400 requests from consuming unbounded disk space by automatically cleaning up old request logs and enforcing a size limit.
  • Fixed unnecessary credential and session updates that could trigger needless authentication reloads in other running processes.
  • Fixed context-overflow recovery for Strata requests that exceed the model context limit but return no usage information.
  • feat: introduce Pro with a minimal one-time desktop card
  • Open in the app: openmausbot://cloud signs in or connects to My Cloud
  • Cloud Pro: included Boat computers and voice as a fallback to the person's own keys
  • [NA] [BE] Update model prices file
  • [OPIK-8466] [FE] fix: read compare panel scores and comments from the live trace
  • [OPIK-8593] [FE] fix: order evaluators in create-experiment snippet to match explanations
  • #14634 34ec938 Thanks @brunoagatao! - Add a copy content button to the Documents viewer toolbar, and expand the document tab context menu with Copy Path, Copy Relative Path, Copy File Name, and…
  • #14630 de285a6 Thanks @brunoagatao! - Add a manual Fetch Models button to the Custom Provider settings dialog.
  • #13589 2d6d0a0 Thanks @sylwester-liljegren! - Suggest files from every folder in a multi-root VS Code workspace when typing @, so folders added through "Add Folder to Workspace..." are mentionable…
  • Add Turkish language support to @openbot/i18n and apps/mobile
  • Find a search phrase that a line break splits
  • Keep Grok's MCP sign-in failure out of provider error toasts
  • npm: bash npm install -g base44@latest
  • Homebrew: bash brew install base44/tap/base44
  • Version: 0.1.25 NPM Tag: latest
Billion Context
  • feat(web): quick-config controls on the config page
  • ci: release canary — automated no-op self-update verification
  • fix: normalize Responses side-request compaction handoffs
  • be01deccaf3dd7b85e03620145bc1a757c857827 .NET: Add origin pinning to Foundry toolbox MCP client
  • 0d6d2bb6b0b58f3aabdf170320ab90fc300d4c99 .NET: fix: declarative workflow http variables
  • 2024d4df4c4dd01b904fb9daaf56f5759801654a .NET: Fix workflow topology edge multiplicity comparison
E2B
  • 2de0cc3: Cap sandbox fork count at 20. e2b sandbox fork --count, JavaScript Sandbox.fork({ count }), and Python Sandbox.fork(count=...) reject a count outside 1–20 before the API call.
  • 03887ad: Apply .dockerignore and fileIgnorePatterns / file_ignore_patterns the way Docker does when copying files into a template, so ignored files are no longer uploaded or included in the files hash
  • A pattern that matches a directory (node_modules, .git, dist/) now excludes everything under it, and a leading / is ignored.
  • InMemoryDocumentStore.bm25_retrieval and InMemoryBM25Retriever with BM25L (the default) or BM25Plus now return only documents that contain at least one query term, with or without scale_score.
  • TextCleaner.run now raises a clear TypeError when texts is not a list or any element is not a str, instead of failing later or producing unexpected results.
  • Haystack now requires anyio>=4.14.2. anyio is installed transitively through httpx and openai; versions before 4.14.2 are affected by CVE-2026-63374 (GHSA-82r6-8w77-94w6).
  • 67d61a2073bad0fbb7140fd2000f1ab7ec0a4f25 MUL-7780 feat(issues): run charts follow the pointer; a cost sparkline replaces the sidebar strip
  • e31da86c90794b5c488279a3ead13ac2f31ac269 MUL-7801 fix(daemon): follow directory junctions when authorizing repo checkout workdirs
  • 262dcb5fa7ee2986aaabf3a78ee698f4afb4f151 MUL-7816 fix(taskfailure): classify a 403 usage limit as provider quota, not auth
  • feat(web): add org-level ingestion overview tRPC route
  • feat(evals): track the evaluator onboarding funnel in PostHog
  • feat(web): Org-level analytics view
  • User Management / Authorization (RBAC): a new agno.os.authz package adds role-based access control to AgentOS — a role store, scope policy, audit log, user directory and an admin router, with…
  • AgentOS Filesystem: a new agno.fs filesystem (DbFileSystem) with dedicated /filesystem routes (list/read/manage files) backed by an agno_fs table.
  • MCP config / auth updates: refreshed MCP server config and built-in MCP auth handling.
  • Tray menu: open or hide the window, see whether an agent is working, see the open project, and quit. The icons follow the light or dark theme.
  • window tool: agents can show or hide the editor window, or check whether it is showing, from MCP or with diffusion window [show|hide]. Every tool works while the window is hidden.
  • Lower memory use: after ten minutes hidden and unused, the window is closed to free its memory. The next tool call opens it again on the same project.
  • <!-- This is an auto-generated description by cubic. --> <a href=" target="_blank" rel="noopener noreferrer" data-no-image-dialog="true"><picture><source media="(prefers-color-scheme: dark)"…
AI Memory
No release notes published.
  • feat: AG-UI 1.0 for CopilotKit (7693a04)
  • fix(runtime): stop reusing per-response provider ids as message ids (5832fff)
  • fix(web-inspector): fit and scale the Inspector on small screens (5135c56)
  • #412 — record which executor ran an iteration
  • #416 — Quick Tunnel DNS negative-cache trap
  • #448 — search truncation accuracy
  • A network failure now says which network failure it was. Node's fetch reports every connection-level failure as fetch failed and keeps the reason on error.cause, which the shared transport()…
  • New connector pair. A CELLxGENE Discover connector discovers public single-cell datasets, and an Alliance of Genome Resources connector covers cross-species model-organism genes.
  • Variant and omics expansion. MaveDB functional scores join the variants connector, and the omics connector expands its Metabolomics Workbench coverage.
  • Library preview Inbox. The workspace library preview gains an Inbox tab for the pending queue, with accept, dismiss, undo, and batch acceptance.
  • Hand back a hosted prompt when the gateway stops answering (6fb8fcd51) — Vincent Hus
  • Install opensre inside the provenance container before recording (70cd2e0ab) — Vincent Hus
  • Stop treating hosted-gateway start and stop as admin-only (527719b8d) — Vincent Hus
Traycer
  • test(gui-app): order F21 manual read-back after automatic completion (72f9380)
  • fix(gui-app): avoid blocking audio setup during startup (6094f5d)
  • fix(gui-app): slide the landing terminal panel instead of resizing it (1a954c9)

Sep 30, 2026

Cline
  • Custom providers added with Add Provider now work when you run a task. They showed up in the provider and model pickers but failed with Unknown or disabled provider
  • New chats remember whether you last used Cloud or Local, and which Cloud model you picked, instead of always going back to Local.
  • Saving provider credentials no longer fails when the provider's model list can't be fetched, and the model list refreshes when you change a provider's API key or endpoint
  • A proxied-server workspace has a backup path. In v1.3.0 bd
  • Fan-in no longer stalls ready work. Closing two blockers of one
  • bd purge can run an orchestrator's wisp retention sweep. bd purge --wisps-plane
Phoenix
  • Updated dependencies [f8ae2fc]
  • @arizeai/phoenix-client@7.16.0
Open Interpreter
  • Fix offline model listing for configured non-OpenAI providers. When an endpoint has no matching catalog identity, the app server now retains that provider's own bundled model choices rather than…
  • Send namespaced session and parent-session identity headers with model requests.
  • Re-sign locally compiled macOS binaries so they run reliably on macOS 27+. (@ryangamerdev)
  • Sign macOS CLI release binaries with a Developer ID.
Caveman
  • First stable release. Requires caveman-sdk 1.2.0 or later
  • caveman_middleware.__version__ is read from the installed distribution.
  • A refused caveman_retrieve (an unknown or expired handle, or a
Experiential
  • A request carrying cache_control on text (block-level or top-level) no longer opts out of automatic Vertex prefix caching.
Omnara
  • feat(machines): add Freestyle provider
  • feat: let models view sandbox images with read_file
  • feat: user and org deletion UI
  • alerts: let the principal find and clear alerts off the live board (7bee3dc)
  • attunement: let people assign a task to a teammate (13da433)
  • knowledge: add the files in chosen Google Drive folders to the knowledge base (354ddb4)
  • [Device connections] Connect over reachable LAN or externally configured VPN networks with IPv4/IPv6 support, automatic address prioritization, and manual selection by network interface.
  • [Messages] Add export to multi-select action bar: export selected messages as Markdown, Word, or to Notion/Yuque/Obsidian/Joplin/Siyuan.
  • [Models] Add GPT-6.1 Sol support for OpenAI and OpenAI Codex, including pricing and reasoning settings.
  • This release tightens the core experience from end to end—faster first turns, clearer chat feedback, sturdier integrations, and a big internal modularity push that sets us up for even quicker…
  • — Thank you @senamakel!
  • — Thank you @senamakel and @M3gA-Mind!
  • fix(xlsx): the sheet help schema declares the set operations the handler applies
  • fix(docx): a style's set places its children in CT_Style / CT_RPr order
  • fix(xlsx): resolve a shared string through a materialized index
  • feat(settings): let Codex terminals opt back into Codex's shared server
  • fix(terminal): run Orca's cmd.exe, path-named and setup-gated Codex launches without the shared server
  • fix(runtime): stop a busy Codex 0.150-0.157 pane reading as tui-idle
No release notes published.
Chat ON Steroids
  • The next message is not held up anymore. Some ChatGPT versions return a long, indented message with its spaces spelled &#x20;.
  • Auto-compaction also works during long-running commands. While ChatGPT kept checking a long command, such as a test run, automatic Compact & resume gave up with "ChatGPT has not confirmed…
  • Easier to see why a stuck chat is not recovered. When the ChatGPT page is not ready for a recovery reload, the app log now says what is holding it, for example a message still waiting for…
  • Feature: enum members are now extracted as nodes with case_of edges in four more languages — Rust enum variants, Zig enum members, C++ enum/enum class enumerators including nested enums, and Scala…
  • Fix: Java calls to inherited methods and super.method() now resolve to the declaring ancestor (walking the inherits chain, nearest declaration wins), instead of dangling.
  • Fix: semantic extraction warns once when a file exceeds the 20,000-character cap and is truncated, instead of silently dropping the tail
  • Updated bundled Claude CLI to version 2.1.286
  • CI: pinned the model for issue triage to fix failures caused by the CLI's new default model
  • MCP servers stay out of the way: servers with the default codemode exposure are no longer listed in the codemode description and no longer block the first prompt.
  • More MCP authentication options: oauth.clientName for servers that only accept known OAuth clients, and "auth": { "provider": "<provider>" } to authenticate HTTP servers with a provider's /login…
  • Anthropic workload identity federation from the Anthropic SDK environment variables. See Use an API key from the environment.
  • FilesystemMiddleware can now store blobs on the backend.
  • 4394bcd release(deepagents): 0.7.21
  • e38e6c9 feat(sdk): allow FilesystemMiddleware to store blobs on backend
  • Added a count such as "2 of 5" to the permission prompt when several permission requests stack up
  • Added mouse support for the "N more" rows of lists in fullscreen mode: click one to jump to that end of the list, with hover and pressed states
  • Fixed several Claude Code processes and IDE extensions each opening a login browser when gcpAuthRefresh or awsAuthRefresh credentials expire
  • Added Factory Droid OAuth, HTTP streaming across Anthropic, OpenAI and Gemini protocols, and pool-aware usage reporting
  • AuthStorage.keys.setConfig(provider, value, { fallback: true }) registers a key that is used only when no stored OAuth or /login credential exists, instead of overriding them.
  • Cursor turns now fall back to Cursor's HTTP/1 streaming transport when HTTP/2 is unavailable, surface Cursor's structured service errors, and resume from the last safe server checkpoint after a…
  • 123 curated skills (language and framework patterns, testing/TDD, code review,
  • Generated deterministically from the explicit include/exclude lists in
  • The build fails on safety violations: non-allowlisted URL hosts, pipe-to-shell
  • fix(cdp): preserve correlated command responses after page replacement #839
  • fix(devtools): unblock navigation while the debugger is paused #703
  • fix(protocol): retain navigation ownership during response body reads #923
nanoMuse
  • Sign in with a phone number (relay 0.5.3). A mainland China mobile number works everywhere an e-mail did — the phone, the web and desktop app, nanomuse.cn/web: the six-digit code arrives by SMS…
  • The face moves on the web and the desktop. The dragon plays the four short clips the phone has — a head shake at rest, headphones and a laptop while working, a crystal ball while waiting, a star…
  • The hub is the one way a computer joins. The local-network host script and Pair over the local network are gone: install nanoMuse Desktop, sign in with the same account, done — on any network.
  • Added an optional model parameter to spawn_agent to choose a model for that spawn instead of using the configured subagent model or inheriting the parent model, with the active model shown in its…
  • Added BYOK support for GPT-6.1 Sol with an OpenAI API key. (#64968; thanks ktKongTong)
  • Added Grok 4.7 to the SuperGrok and xAI providers and made it their recommended default. (#64567; thanks raphaelluethy)
  • Cloud workspaces and tasks become shared records: presence, sharing, descriptions, comments, labels, mentions
  • chore(desktop): bump version to 1.32.0
  • feat(mobile): one bottom sheet for page comments, not a floating card
  • Subagent research files land in the workspace. When a workflow gave a subagent a relative output path such as outputs/.drafts/<slug>-research-web.md, pi-subagents saved it under…
  • feynman packages remove <preset> (also rm and uninstall) removes an optional package such as memory.
  • On Windows, Feynman no longer fails to start when another Feynman start is copying keybindings.json (EBUSY: resource busy or locked).
Kortix
  • Computers are connector accounts like any other: the same row, the same Share dialog (people, groups or everyone), and one registration per machine.
  • The audit trail shows Via for each event: the credential the API authenticated (browser session, personal access token, connected app, session token, API key, service account or SCIM token), never…
  • Slack and Teams connector reads stay inside the calling project's own conversations.
  • fix(react-core): use new run IDs for standard interrupt resumes (081e745)
  • fix(vue): stop cloning run state six times per row for message slots (bd39b04)
  • feat(web): add Mod+N / Mod+Alt+N shortcut to create a workspace
  • fix(web): keep the workbench stable across workspace switches
  • fix(web): stop remounting the workbench after a panel resize
  • 94ef1598fff72b824b1d5d9bfe727529332717d1: chore(electron): upgrade Electron from 42.3.0 to 42.11.8 for Theia v1.74.2
Billion Context
  • fix: report unrouted-endpoint direct sends for POST requests only
  • fix: bound the native fetch chain against coexisting re-wrapping plugins
  • fix: classify opencode v2 title-gen by host-declared intent, not token budget
  • Export cancel: cancelling an export now asks for confirmation. The progress view also says whether it is exporting audio or video.
  • Animations: overlapping animations now combine instead of overriding each other, and presets adjust the node's own properties instead of replacing them. Fixed text motion overrides on static text.
  • Captions: the Guinea and Spotlight caption styles use Outfit and have a new highlight color.
  • fix(engine): remove per-sandbox piece LRU @amrdb
  • session.respondToToolApproval(...) now requires toolCallId (id-less approval responses are rejected); related approval APIs now key gates per thread/run and update grant scoping.
  • @mastra/playground-ui removed/renamed multiple UI APIs (notably ScrollArea props, PageHeader props, ToolCall → Activity component renames, and significant color token API changes).
  • Added maxRetries and failurePolicy options to the observational memory observation and reflection settings.
  • [OPIK-7532] [QA] chore: reconcile coverage taxonomy with spec tags
  • [OPIK-8591] [FE] fix: show selected-columns count on Experiments page column picker
  • [NA] [DOCS] Quarkus LangChain4j: enable prompt and completion tracing
  • feat(web): onboard self-hosted clients and integrate OOMOL Key
  • fix: treat blank number-like values as missing
  • fix: refine OOMOL Key visibility and Marketplace errors
  • feat(seeder): add incident-session scenario for the session timeline
  • feat(skills): track draft changes and fetch file contents by hash
  • fix(web): reword AI features data notice to "used to train models"
  • Fix Enter, Backspace and arrow keys in the browser live view
  • Make the Computer Use result copy smaller for OpenCode
  • Stop leftover driver and provider processes, and cut idle memory
  • c9adca039f3c24b570567e0def8d1018e9194c9f feat: add Audit provider to forward audit logs to Syslog
  • 99447318f951724b9d538e0fd025cbc9a31bd9f8 fix: keep provider host, endpoint and type unchanged on demo database
  • 11f53fbf9f2f53de650c7451106a72c2f8b63ac5 fix: publish all certs in the global JWKS again
  • libfx now uses LIBFX_MODEL_UNSUPPORTED_EFFORT and LIBFX_MODEL_UNSUPPORTED_FAST for unsupported model settings. Callers checking the old error codes must update.
  • libfx accepts model: { id, effort, fast } and supports Blob or File images directly.
  • Detached processes such as tmux, ssh-agent, and browser drivers now stay alive between shell calls.
  • Quota burndown: Plan Usage now shows the recorded remaining-quota burndown for Codex and Claude next to utilization history, with capture age and calendar endpoints. Thanks @callmejustdodo!
  • Refreshed provider colors: 16 accents now match official brand sources, while colors that would hurt menu bar or widget readability keep their current values. Thanks @elijahfriedman!
  • More accurate costs: Mistral usage is priced by event type, API zone, and service tier, Antigravity and Codex model aliases and published Cyber rates are priced, and Mistral's Monthly Plan can…
  • Release notes — formatted for people, with expandable sections.
  • Changelog — plain Markdown for AI agents and tools.
  • npm package
BrowserSkill
  • fix(geometry): unify cross-frame coordinates and screenshot conversion | 统一跨 frame 坐标与截图转换
  • refactor(observation): decouple DOM processing from capture orchestration | 解耦 DOM 处理与采集流程
  • refactor(observation): unify cross-frame capture coordination and data ownership | 统一跨 frame 采集协调与数据归属
  • Added the Queued Messages option to /config: keep the default Queue to hold follow-up messages until the current turn ends, or choose Send Immediately to interrupt the agent with them.
  • Improved remote-control start on Linux machines without a systemd user service manager, such as most containers: instead of failing, it now starts the daemon as a background process and warns that…
  • Improved loading long conversations: resuming a conversation with thousands of steps is noticeably faster because the CLI no longer scans every step up front
  • TTSE dual-track online self-evolution – Added a TTSE (Two-Track Self-Evolution) framework for online self-evolution across environment facts and task experiences, with evolution signal detection…
  • Proactive Context Service (PCS) – Added multi-source context construction and runtime injection to JiuwenSwarm.
  • Swarm Evolution self-evolution capabilities – Expanded the framework for autonomous capability evolution.
  • Skill installation: install from Cow Skill Hub, GitHub, or ClawHub, or upload an archive, a SKILL.md, or a folder. Every skill is previewed first and only installed after you confirm.
  • MCP configuration: add, edit, and remove MCP servers in the UI, covering the stdio, SSE, and Streamable HTTP transports.
  • Tool retrieval diagnostics: see how MCP tools were ranked and why retrieval fell back during a conversation, making it easier to find out why a tool was not picked. Thanks @zhongjunqi7657
  • test(windows): verify native shell_run end to end (e81c88df1) — Anwesh
  • fix(skills): resume CI demo helpers after gateway task replacement; drop separate failure check (194301c27) — Vincent Hus
  • docs: end the public-alpha note with a period (f1cecde76) — Vaibhav Upreti
  • 内置账号登录:通过 /provider 或 /auth 管理 ChatGPT/Codex、Claude、Grok 订阅登录;在 DSH 0.2.0-rc.1+ 上可完成 DeepSeek 浏览器授权,再通过 /model 选择独立的 deepseek-account 路由。标准 profile 无需另装 dsh-auth,原 OAuth 凭据继续可用。
  • 实时任务面板:在支持内核事件与输出环的版本上,/jobs 实时显示任务进度、stdout/stderr 和输出丢失提示;同时修复有主任务无法列出或终止的问题。
  • 随包 AI 手册:默认通过 dsh-tui 或 dst 启动后,可以直接问 AI 本工具的用法,也可以运行 /dsh-tui-guide <问题>;手册正文仅在需要时加载。
Pydantic AI
  • GHSA-v36g-jcw9-x7cw (moderate): converting attacker-controlled HTML with deeply nested elements in the local web_fetch tool could consume excessive CPU and memory.
  • Cap genai-prices below 0.1 to keep token usage extraction and limits working

Sep 29, 2026

  • Fixed OpenCode v2 turns longer than five minutes failing with UND_ERR_HEADERS_TIMEOUT
  • Fixed the context meter staying empty during OpenCode v2 turns and disappearing after them
  • Fixed OpenCode v2 question cards showing only the header and offering no typed answer
Omnara
  • feat: Skill creation ux and fixes
  • fix: update signal and pulse examples for the current SDK event stream
  • fix: issues where MCP returns a 500 instead of a useful error code
Traycer
  • test(desktop): stop a replayed FSEvents edge racing the hysteresis test (0433ba4)
  • fix(desktop): make the pid.json start-gap test deterministic, log lost watcher edges (4e49430)
  • protocol: sync released baseline surface fixture (2b431c3)
  • fix(desktop): correct Grok Build setup guide URL (7ebf1beb44139264e9c5c7fee6970429ff66f934)
  • feat(relay): implement NIP-AR channel artifacts (12670bd0f037c66a682272bb81c46c3f254fad74)
  • fix(desktop): resolve unlisted project channel requests (c96553eb13a073ca9f4ee1547b50ba3d28dfd4db)
  • briefing: feed the principal's live day into the header and the brief (7fea2d5)
  • briefing: ground names and figures in unattended prose before delivery (a565c30)
  • memory: keep chat corrections as standing facts in prompts (4e2a42b)
Experiential
  • Accept service_tier: "fast" as the canonical Priority processing tier on Chat Completions and Responses.
  • Preserve complete tier-specific long-context and cache-write price schedules.
  • Reserve the larger of the frozen Standard and requested-tier schedules, then price the actual served tier from provider evidence.
  • Fixed Codex subscription model discovery to claim the current stable Codex CLI release (0.159.0, up from 0.153.4): ChatGPT gates the discovery endpoint on client_version, and the stale pin hid…
  • Fixed the model picker showing a stale model list for up to a minute when opened without a search term (Ctrl+L or plain /model); it now always refreshes the catalog on open
Feynman
  • Pi 0.99.1, pinned. Feynman accepted any Pi version, so npm installs picked up Pi 0.99.1 the day it was released, untested, and Pi 0.99 warns on every launch about how Feynman declared it.
  • npm installs use a patched undici. pi-subagents pins undici 8.10.0, which ten advisories published on 2026-09-29 affect (plus GHSA-3wwx-pv8p-q78v), and npm hoisted that copy for every package in…
  • feynman model set explains a model it cannot use. "Model not available in Pi auth storage" gave no reason.
  • Embedding fixes prevent silent truncation and size chunks by each model's token limit so your content is embedded intact and consistently.
  • Slack integration: import and continuously sync conversation history from Slack into Cognee.
  • Skills moved to a new .agents/skills location with Windows linking support and recovery when linking fails.
  • feat(angular)!: render A2UI without Lit and support web component cat… (b3d1aad)
  • fix(core): distinguish Intelligence replay errors from live failures (278a78d)
  • fix(mcp-apps-renderer): follow the host page's dark theme (2476de1)
  • fix(a2a-server): add early return on unsupported store in tasks metadata endpoint
  • Changelog for v0.61.0-preview.0
  • Changelog for v0.60.0
nanoMuse
  • nanoMuse Web — nothing to install. Sign in at nanomuse.cn/web with an e-mail or a phone code and you get a nanoMuse of your own on the project's server: kept between visits…
  • nanoMuse for the desktop — the window. A desktop app for Windows, macOS and Linux shaped after the phone app, with the runtime inside: a face, one main chat and side chats, Devices, approvals in…
  • The computer on the hub, both ways. nanomuse serve signs in to nanoMuse Cloud, answers the account's other devices…
  • BedrockConfig(session=...) takes an aiobotocore.session.AioSession. A boto3.Session fails on the first call with AttributeError: 'Session' object has no attribute 'create_client'.
  • The bedrock extra installs aiobotocore>=3.9.1,<4 instead of boto3. If your application also installs boto3, pick a version whose botocore fits aiobotocore's pin.
  • Explicit keys and region_name passed together with session= now apply. They used to be silently ignored.
  • Google extensions (Gmail, Google Calendar) can now be activated on a
  • The WASM runtime now uses Wasmtime and WASI 48.0.3, fixing recent guest fuel
  • 🗂️ A unified workspace browser and compact file panel with badges for added, modified, and deleted files, plus desktop niceties: Show in Finder for local files, an opt-in setting to open chat…
  • 📦 Sandbox upgrades: Linux agents can make disposable edits with copy_on_write: true in write_paths, admins can opt into shallow or blobless repository clones, and Daytona sessions keep their…
  • 🔐 Admin controls: default new sessions to public (off / sandbox / all) from Settings > Sharing, trust whole subdomains with *.
  • fix: reject mixed blocked DNS answers with IP allowlists
  • fix(agent): compute recursion_limit accounting for middleware graph nodes
  • fix(ci): keep PR filenames out of shell scripts
  • Updated bundled Claude CLI to version 2.1.285
Kortix
  • Local mode: connect your computer from the desktop app. Each computer is a connector account with access approval, and the tunnel keeps running in the background while the app is closed.
  • Microsoft Teams reaches Slack parity: direct messages, a connected note, a home tab, message actions, edit and delete, /unbind, /status buttons, /sessions, the full agent picker, review cards, and…
  • Saved session history is always on, and a reload shows the newer saved copy.
jcode
  • jcode is now a first-class agent in herdr: its name and working/idle state show in the sidebar and herdr agent list, herdr agent wait works on it, and after a herdr restart the pane reopens the…
  • /cloud and /local move a live session, including uncommitted code, to a cloud host and back
  • herdr reports run in the background with a short timeout, so herdr can never slow jcode down, and jcode releases the pane when you quit
OpenMausBot
  • feat: introduce Pro with a minimal one-time desktop card
  • Open in the app: openmausbot://cloud signs in or connects to My Cloud
  • Cloud Pro: included Boat computers and voice as a fallback to the person's own keys
  • fix(terminal): run Codex in Orca terminals without the shared background server
  • fix(daemon): roll Codex no-daemon shell launch into a fresh v37 daemon
  • fix(runtime): settle a quiet Codex composer as ready on every version
Oh My Pi
  • Added Agent.replaceQueue() to replace one pending queue without changing the other queue
  • Added queued-message grouping so owned companion records and their user prompt are dequeued together in one-at-a-time mode
  • Added Agent.onQueueChange(), a listener called whenever a steering/follow-up queue mutator (enqueue, dequeue on delivery, clear, or restore) runs, so hosts can observe queue changes without polling
Herdr
  • Terminal shortcuts that send Escape followed by a key work again in panes. On macOS, Option+Left/Right and Option+Backspace from Ghostty's defaults or iTerm2's Natural Text Editing preset move and…
  • Alt+[ followed quickly by another key no longer merges into a different key.
  • Added CLAUDE_CODE_DISABLE_WEB_FETCH environment variable to turn off the WebFetch tool
  • Added claude --desktop to open the Claude desktop app on the current directory, or on a session with --continue / --resume <id>
  • Added claude plugin configure <plugin> to show a plugin's options and which are unset, or save new values read from stdin with --values-stdin
  • Build: 0.12.6 Commit: fa700c113e Built at: 2026-09-29 19:21:38 UTC
Phoenix
  • Updated dependencies [5b37f12]
  • @arizeai/phoenix-client@7.15.0
Sim
  • feat(library): Best ChatGPT alternatives for building AI agents and automating workflows
  • feat(library): Best AI Workflow Builders for Small Teams in 2026
  • feat(forks): opt-in fork sync for new workflows
Openbot
  • Let every member import agents on a joined server
  • Show the Dynamic Island state as an iOS Live Activity, updated through APNs
  • release: prepare mobile v1.1.0
  • Feature: after a package upgrade, graphify refreshes stale installed skills automatically (the SKILL.md + references sidecar it manages) so the version-mismatch warning no longer requires a manual…
  • Fix: graph.html's Node Info panel now shows the real Type/Source/Community for each node instead of "Type: unknown / Source: -"…
  • Fix: a Kotlin class property that is both annotated and has an inferred type (@Volatile var x = 0) no longer crashes extraction with an UnboundLocalError that dropped the whole file
Pi
  • GPT-6.1 Sol — Available on OpenAI, Azure OpenAI, and OpenAI Codex, and now the default OpenAI Codex model. See Select a model.
  • Added GPT-6.1 Sol (gpt-6.1-sol) to the OpenAI, Azure OpenAI Responses, and OpenAI Codex providers.
  • Changed the default OpenAI Codex model to GPT-6.1 Sol (gpt-6.1-sol).
  • 9f1300532a7fbffe0d7d39fdde1c098403cfd970 feat: add application-level token group format option
  • 3377f35609b84c54d4bea0fd5b48c54a70948abd feat: support OIDC prompt=none on the authorize page
  • de6fc6d2621e835ca10dbee0a99866921ba04a50 feat: support RFC 9207 iss parameter in authorization responses
No release notes published.
  • The internal golden carries psql, ntn, lim and the Stripe CLI
  • fix(slack): open page details in the Work Object flexpane
  • feat(cli): coworkers set up their own cloud environment from the CLI
Deep Agents
  • MCP tool calls now use configurable timeouts to prevent them from hanging indefinitely.
  • Stdio MCP servers now start in the session directory.
  • JavaScript subagent costs are preserved when resuming a session.
Billion Context
  • fix: narrow isCodexClient fallback from substring to token-level prefix
  • fix: hold SSE clients across streaming silence past undici 300s body timeout
  • test: widen stall-guard stream-test budget margin to kill CI-load flake
  • feat(allowlist): add Jinja template support
  • fix(idea): list merge commit files against the first parent
  • fix(make): append .exe to the dev binary on Windows
Opik
  • [OPIK-8343] [BE] fix: bound the experiment reads of traces to the weeks their items resolve to
  • [OPIK-7596] [BE] fix: hide internal workspaces from the MCP OAuth consent list
  • [NA] [SDK] [DOCS] Update automatically OpenAPI spec and Fern code
OpenSRE
  • fix(analytics): refresh identity after external account changes (800220b51) — Vincent Hus
  • fix(analytics): correlate CLI browser authentication attempts (4b183f3a2) — Vincent Hus
  • fix: close the open CodeQL security and quality findings (c3799af11) — vaibhav upreti
  • Godot 4.2+
  • Active Ziva account
Crush
  • f2d66d286950109082c102d780013df84f741e55: fix: remove the powerline symbol (@andrinoff)
  • fix(v2): single cache breakpoint, internal wakes, boards in summaries
  • refactor(task-session-manager): simplify reopen correction delivery
  • fix(orchestrator-wake): failed sends must not stall idle parents
  • docs: changelog and blog post for v0.10.1
  • docs: every Hindsight code example comes from a tested snippet (+4 fixes it surfaced)
  • blog: What's New in Hindsight Cloud (0.10.0), plus three stale-line fixes
  • Octop↔Octop 云端桥接,经隧道使用远程专家
  • 内置助手可回答产品与使用问题
  • 可配置 Ollama 本地模型下载目录
  • Object mask tool: pick an object in the viewer with points or a box, then refine the mask with a brush.
  • Clip paths: the old "mask" feature is now called clip path. It has its own toolbar tool, keyboard shortcuts and inspector section, and you can start a clip path directly from the inspector.
  • Tracking settings: the inspector has tracking options for shapes.
  • 为 ls 补充目录摘要和概览控制参数
  • 新增 ListPage 和 TreePage,支持获取 has_more
  • 保持原有 List 和 Tree 接口向后兼容
  • Stock profiles carry the listing's own identity
  • Volume statistics in technical_indicators
  • Argentina (BYMA) market data
  • feat(topics): render and measure dedicated trace transcripts
  • feat(native): prepare JS events and encode batches through NAPI
  • feat(observability): label experiment query outcomes for SLOs
Base44
  • npm: bash npm install -g base44@latest
  • Homebrew: bash brew install base44/tap/base44
  • Version: 0.1.24 NPM Tag: latest
  • 💻 Windows:双击下载的安装包,按提示完成安装。
  • 🍎 Mac:打开下载的安装包,将 toonflow.app 拖入“应用程序”,再从“应用程序”打开。
Glimmervoid
  • Team reviews can build the pull request: the reviewer now builds and runs tests in a copy of the pull request inside its own review folder, with your clone's node_modules linked in read-only, so…
  • commit: 2a0ccfcd4ca63d2c91ffe9dad4f40fad63037f2b
  • source tag: beta2026-09-29_10-30-38
  • regions: cn global
  • Parallel analysts. The selected analysts run at the same time, so the analyst phase takes about as long as the slowest one.
  • Unattended CLI runs. Flags answer the per-run questions, and TRADINGAGENTS_* variables the rest, so a scheduled job or script asks nothing: tradingagents --ticker NVDA --date 2026-09-26 --analysts…
  • Reports record what produced them. Each report opens with the analysis date, version, provider, models, analysts, rounds and data vendors; the state log keeps them as run_settings.
  • The hub now starts reliably on Windows machines with a system proxy. Bun's fetch sends loopback requests through HTTP(S)_PROXY with no localhost bypass, so on machines that export those variables…
  • The hub daemon log now records one line per socket close with the close code and reason, whether the heartbeat sweep dropped it for a missed pong, and whether the server was already shutting down.
  • Token usage no longer counts reasoning tokens twice. Providers report reasoning tokens as a subset of outputTokens, but normalizeUsage() returned the raw total alongside reasoningTokenCount, so…
  • 透明终端图像:LaTeX 公式渲染成真图(Kitty/Sixel),默认透明底透出终端壁纸,点击放大 2 倍预览;块级公式可放大(mathImageScale);标题女仆娘立绘与聊天照片同样透明,底色每面可配(mathImageBacking / imageBacking)(#1095 #1098 #1103 #1107 #1149)
  • Resume 提速:物理指纹复用会话摘要 + 跨重启首屏快照,千级会话秒开(#1140)
  • /provider:模型列表实拉端点并与内置目录合并(#1141)
  • Improved rate-limit handling when the model API returns a retry delay: the CLI now waits the delay the server asks for instead of a fixed 5 seconds, and stops right away instead of retrying when…
  • Improved rendering efficiency, cutting CPU use and memory churn while the conversation view redraws, such as during fast scrolling or while the agent is working
  • Improved the artifact viewer's m hint to name the content and the mode it switches to, such as diagram ASCII, diagram source, or math image, instead of toggle ASCII or toggle raw
  • Current flagship models: add Meta Muse Spark 1.3, Anthropic Fable 5.1, OpenAI GPT-6 Astra, and OpenAI/fal GPT Image 2.5 support to the extended-stable line.
  • Complete GPT-5.6 family support: preserve Sol, Terra, and Luna catalog, routing, migration, image, vision, and thinking behavior, including Ultra reasoning across model-runtime boundaries and…
  • Extended-stable security rollup: reconcile all repository advisories that affect 2026.8.2, harden Prometheus metrics authorization and Discord asset/voice ownership, and clear the production…
  • Support saving and restoring new window layouts
  • Support double-clicking parent document titles to open them
  • Support expanding list mind maps by level
  • M3gA-Mind — thank you!
  • senamakel — thank you!
  • siddarthkay — thank you!

Sep 28, 2026

  • Native Linux ARM64. Open Science now ships ARM64 installers for Linux alongside x64.
  • Connector expansion. A new Pathway Commons connector joins expanded cBioPortal, openFDA, MGnify, and Bgee cross-species expression sources.
  • Journal datasets. The literature library gains journal datasets with reference attributes for organizing collections.
  • Changed the continual-harness kernel surface back to the twelve per-kind CRUD methods on rlm.harness…
  • Fixed harness state persisted while the grouping was spelled topic losing that grouping after the revert: entries migrate to path when they load, and rolling back a refinement recorded in that…
  • Fixed the conversation detail level resetting to Details every time you open, resume, or attach to a chat: the level you pick with Ctrl+O is now saved as the chatDetail setting and reused for…
  • Ask isolated side questions with /btw.
  • Offer to summarize threads and switch when the cache expires.
  • Persist project MCP startup denials.
Paseo
  • Added Claude Sonnet 5.5 for Claude Code 2.1.284 and newer
  • Fixed OpenCode chats failing with "Variant unavailable" after switching to a model without the selected thinking level
  • Fixed rewinding a Codex chat dropping its custom provider and Paseo tools
  • #938 efe6f0f Thanks @SammyTourani! - fix: --debug now shows the innermost cause of a failed run, such as the network error behind "Connection error."
  • #937 261022d Thanks @ind1go! - Correct MCP generation for IBM Bob.
  • #840 d64ac87 Thanks @Yashwanth-Kumar-Kotla! - Fix .last-update.json being written non-atomically, which could leave it truncated/corrupted under a failed or concurrent write and silently discard…
Experiential
  • Exact Sonnet 5.5 metadata with high default effort, 1M context, 128K synchronous output and published base/cache pricing.
  • Native Messages support for type-only thinking: {"type":"between_tools"} at low, medium and high effort. Invalid modes, fields and effort combinations fail before dispatch.
  • Preserve between-tools semantics through reasoning-channel resolution and refuse unsupported transports rather than substituting a different thinking mode.
  • Add support for native Gemini 3.8 Flash
  • Implement evaluation of the last traced run through AMP in crewai eval
  • Record the last traced run for crewai eval instead of printing it
  • otel: fix register() crash with opentelemetry-exporter-otlp-proto-http 1.45 (70c5051)
  • Changes since langchain==1.4.2
  • release(langchain): 1.4.3 fix(langchain): sanitize cache settings for fallback models feat(langchain): support Bedrock Mantle chat models in init_chat_model fix(langchain): recognize GPT-6…
  • feat(react-core): add transformMessages to CopilotChatMessageView (eb460b3)
  • fix(core): keep the reconnect cursor when a control frame repeats an old checkpoint (d3c36bb)
  • fix(runtime): enforce basePath segment boundary in single-route mode (15437bb)
  • Fix status bar on smaller screen
Chat ON Steroids
  • OpenRouter keys work even when pasted with hidden characters. A key copied together with an invisible character was sent unchanged, and OpenRouter rejected it with "Missing Authentication header".
  • Replies that point to earlier content show properly. Sometimes ChatGPT answers by pointing to content from another message, for example when you send the same short message again.
  • New ChatGPT formats no longer appear as code. ChatGPT keeps adding special formats to its replies, different per account.
  • Updated bundled Claude CLI to version 2.1.284
  • fix(marketing): hide leaderboard vertical guide lines
  • fix(desktop): restore locale provider test mocks
  • chore(desktop): bump version to 1.30.2
  • Plugins feel native: user plugins now get their own switcher tab by default, and Notion AI, ZoomMate, and LongCat run as bundled plugins with browser sessions kept private to the host.
  • Lighter on CPU and disk: Claude and Vertex cost history is reused instead of re-decoded on every scan, the history cache is about a quarter smaller, and cost scans no longer expand priority days…
  • Steadier refreshes: Codex rereads credentials while the CLI rewrites them and picks up plan upgrades right away, a stalled Keychain signature check can no longer freeze every provider, and Claude…
BeeAI Framework
  • tools: generate typed schemas for keyword-only parameters
  • propagate Python stream errors and cancellation
  • adapters: preserve LangChain image content
  • Added Claude Sonnet 5.5 (claude-sonnet-5-5), now the default Sonnet model on the Anthropic API — 1M context, $2/$10 per Mtok with $0.20/Mtok cache reads
  • Added a "Yes, but ask again next time" answer to auto mode's prompt before a read outside the working directories, so you can allow that one read and still be asked about later ones
  • Added dollar amounts to the Claude apps gateway spend limit in /usage and the status line (for example "$271.40 / $500.00 spent this month") when the gateway runs this version or later.
Traycer
  • feat(antigravity): harness protocol ids and GUI registration (1139e0d)
  • fix(gui): open an epic on its most recently active agent, not its oldest node (e668529)
  • fix(gui-app/mobile): disable the tab-switcher trigger until a sheet is mounted (3529b18)
  • Keep the reason of a failed turn and stop two false errors
  • Detect local providers, edit endpoints, and add custom ACP agents
  • Fill the full row with the selected server highlight on mobile
  • delegation: draft replies in your voice from your own Gmail (cc2e95f)
  • knowledge: read scanned PDFs on every channel (a083f18)
  • providers: read PDFs through each provider's own file support (9e0bbc4)
  • A timeline has a colour: SDR (Rec. 709), HDR (HLG) or HDR (PQ), set in
  • HDR clips - an iPhone's HLG, HDR10's PQ - are decoded at ten bits and
  • On an HDR timeline the clips keep their light above white. The monitor
Opik
  • [DND-1745] [BE] Bind 'unknown' instead of NULL for the span/trace source column
  • [OPIK-7791] [QA] e2e specs for release 2.2.79
  • [OPIK-8365] [BE] perf: replace spans FINAL with a latest-version dedup in GET_TOTAL_COST
Feynman
  • OpenCode no longer recommends a retired Kimi model. OpenCode Go and OpenCode Zen retired kimi-k2.6 ("Use kimi-k2.7-code instead"), but Feynman still picked it as the default for new OpenCode users.
  • A failed literature database request now names the service, for example www.ebi.ac.uk request failed: 503, instead of "Science database request failed".
  • The PageIndex SDK, local or cloud — vectorless, reasoning-based RAG, end to end.
  • Much faster indexing — the PageIndex Flash engine gets the tree from layout stats: no LLM involved for the structure generation itself, LLMs only write the node summaries, and tree expansion…
  • Flash engine: the local default (mode="standard" keeps the classic LLM pipeline).
Billion Context
  • fix: fold round-2 renders the steady path's per-wire shape
  • release v0.1.169
  • Docs: large multi-section documents open without pinning a CPU core, and a resource watchdog offers to close a runaway document; table column widths preview live while dragging a border (reported
  • Sheets: Excel key chords for number formats, borders, Insert/Delete Cells (Ctrl+Shift++ / Ctrl+-) and Format Painter lock.
  • Slides: create --type pptx renders previews and audits the deck, and deck shapes support flipH/flipV (@x1a0Y4NGren); cancelled deck runs no longer report their page step as done.
Oh My Pi
  • Added tool_execution_end events that fire as each tool call settles for live UI updates
  • Emitted tool result messages in the order of tool calls, preserving call order regardless of completion order
  • Reduced repeated token-counting work with a bounded, model-scoped cache of exact text and short-message fragment counts.
  • npm: bash npm install -g base44@latest
  • Homebrew: bash brew install base44/tap/base44
  • Version: 0.1.22 NPM Tag: latest
  • Feature: SQL CREATE TRIGGER statements are now extracted and linked to their table (ON <table>), including OR REPLACE/OR ALTER, INSTEAD OF, and procedural BEGIN…END bodies that previously landed…
  • Feature: Groovy enum declarations and their constants are extracted, with members linked to the enum via case_of
  • Fix: R class definitions created via a namespace-qualified constructor (R6::R6Class, methods::setRefClass) are now recognised, so the class body and its methods are no longer dropped
  • One local service, three interfaces. deepcode, deepcode desktop, and the new deepcode web all connect to the same local background service. Closing a client leaves running work in the background.
  • Dictate your prompt. Add a dictation block to your config and the Desktop and Web composer gains a microphone.
  • More providers, truer numbers. Opper (one EU-hosted, OpenAI-compatible key for many vendors) and Amazon Bedrock join the templates, and DeepSeek V4 models use their own 1M-token window and pricing…
  • commit: da94e838ee58310f8ba015c1657a4bd014ee897a
  • source tag: beta2026-09-28_12-48-19
  • regions: cn global
  • feat(llm): exercise a tool-call round trip in ocr llm test
  • feat(llm): add OpenRouter to list of providers
  • feat(session): add 'ocr session rm' to delete a saved session
  • dbfc014bdd0f0052e732bf4ac63a18084ebef924 MUL-7610: fix(autopilot): avoid loading task history during webhook recovery
  • 7dac88a6a8c5b2e28b3b57d83142722163b24cd1 MUL-7649: issue deliverables and dynamic blocks — sidebar section, overview grid, viewer info panel, html/mermaid block frame
  • 6dc6a0b9a2db36ac31995c7a98ec2d36a1c9f188 MUL-7650: attachment viewer — CSV tables, JSON/YAML tree, numbered code, HTML viewports, every kind in a new tab
  • feat(chat): add shared find menu
  • feat(search): expand live providers and discussion reads
  • feat(snowflake): add Cortex Analyst operations to the Snowflake block
  • 品牌图标焕新为透明底 firelumi mark;默认视频模型改为 MiniMax(Seedance 垫底);修复终端启动管道关闭导致主进程 EPIPE 崩溃
  • 运行时配置与 Account 隔离:新增 Cluster / Account 两级运行时配置(GET/PATCH /api/v1/admin/configuration、/api/v1/admin/accounts/{account_id}/configuration),无需改 ov.conf 或重启;Account 可独立配置 VLM、Query Planner、Embedding、远端…
  • Skill 检索与分发:Skill 整包参与索引,skills/find 与 search(mode="context") 每个 Skill 只返回一条最佳命中;MCP 新增 add_skill 工具;各 memory 插件在会话开始注入 Skill 目录并附带 openviking-skills skill。
  • 检索与存储:本地向量引擎 cosine 分数归一化到 [0,1];新增 openGauss DataVec 向量后端和 Jev rerank provider;grep 支持匹配行上下文,fallback 目录遍历分页,canonical session 使用原生扫描;增量资源导入优化;统一存储 URI 规范化,兼容中文与含空格路径;tags 新增…
  • Standard Operating Procedures (Experimental): Added the agentscope.sop module for splitting a task into ordered milestones, each verified before the next one starts, with run state that can be…
  • Team Pipeline: Added a TeamPipeline class where a leader agent delegates tasks to member agents, which run concurrently and hand their results back.
  • More Realtime Voice Models: Added OpenAI Realtime, Gemini Live, and xAI Grok Voice models to the realtime voice agent.
  • Cloudflare AI Gateway models now honor provider response and stream timeouts. (@danlapid)
  • MCP browser launch failures are now reported when the launcher exits immediately.
  • Debug configuration output now redacts credentials and sensitive headers.
  • docs(policy): clarify sandbox-local loopback access
  • docs: refresh architecture and extensibility pages
  • fix(supervisor-network): keep workload bytes read with a mediated CONNECT header
  • feat(wisburg): add Wisburg provider
  • fix(mcp): describe tool contracts and correct stale agent guidance
  • feat(providers): expand provider capabilities and align runtime behavior
  • Built-in voice input for the CLI, with native microphone capture on macOS and Windows and coding-aware dictation vocabulary
  • The session status line (directory, branch, git status, context, provider, and model) stays pinned below the input, with info widgets showing the detail behind it
  • Agents can show interactive applets, and swarms can message each other across swarms with unique labels
  • A session whose sandbox name was still held by an old archived sandbox could never get a new sandbox. It now provisions under a fresh name.
  • Opening a session repairs a stale runtime immediately instead of waiting for the background check.
  • A prompt parked while the sandbox restarts keeps its sandbox deadline through every retry.
  • #1423 316538a Thanks @jiangjiang248! - Allow custom workspace sandboxes to receive the original execute_command command and args by disabling toolkit command normalization. Fixes #1422.
  • fix(analytics): keep one session id per CLI process (346909af9) — davincios

Releases by week

Every stable release of the week, Monday to Sunday, for the last 26 weeks. Each agent's full history is on its own page.