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: Aug 24 – Aug 30, 2026

286 stable releases from 108 tracked agents and frameworks in the week of Aug 24, 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.

Aug 30, 2026

jcode
  • Internal reliability and performance work
  • Linux x86_64: available
  • Linux aarch64: available
AI Memory
No release notes published.
Agno
  • KnowledgeManagementTools constructor flags renamed, ingest_path now opt-in: the 2.x-style enable_ingest / enable_remove flags are replaced by flags named after the tools they register —…
  • agno.tools.knowledge_management import path moved: KnowledgeManagementTools (introduced in 3.0.3) now lives at agno.tools.knowledge; the old module path is gone.
  • agno.tools.file and agno.tools.knowledge are packages: FileTools and FileGenerationTools share agno.tools.file.
  • GitHub App webhooks now require a configured secret: get_body() no longer accepts unsigned webhooks — a deployment without GITHUB.WEBHOOK_SECRET set is rejected with HTTP 403 instead of trusting…
  • Dependency management migrated to uv: requirements.txt, requirements-dev.txt, and the legacy setup.py are gone — dependencies now live in pyproject.toml with a committed uv.lock.
  • Linting consolidated on Ruff: flake8 and the standalone isort pre-commit hook were removed. Contributors should run uv run ruff check --fix on touched files before committing.
  • ChatGPT ↔ Codex planning and review loop
  • Structured INIT → PLAN → EXECUTED → REVIEW → DONE workflow
  • ChatGPT independently inspects workspace state, diffs, tests, and execution summaries
Casdoor
  • add missing organization segment to syncer add URL (54df9cc)
  • Fix: a batch of cross-language inheritance-edge corrections (thanks @Synvoya): JavaScript class X extends Y now emits an inherits edge.
  • Feature: Robot Framework .robot/.resource files are now extracted (optional [robot] extra) — suites, test cases, user keywords, keyword-call edges, and resource/library imports, with…
  • Fix: chat-template control tokens are now defanged by form (<|…|>, [INST]/[SYSTEM]) rather than an enumerated few, closing a prompt-injection gap for attacker-chosen tokens (e.g. <|eot_id|>).
  • Slides: right-to-left support — paragraph and table direction toggles, mirrored bullets, complex-script shaping.
  • Slides: chart fidelity — manual plot layouts, RTL-aware legends, theme-override colors.
  • Sheets: million-cell copies twice as fast with live formulas; edits that would break formulas or oversized files are rejected up front.
  • npm: bash npm install -g base44@latest
  • Homebrew: bash brew install base44/tap/base44
  • Version: 0.1.12 NPM Tag: latest
  • Read papers with the agent. Link up to three PDFs to a session as reading context — the agent can read the current page, page through the whole document, and search across them.
  • Notebook variables at hand. The shared terminal suggests live kernel variable names as you type, and on wide previews the live Variables pane docks beside cells and the terminal instead of…
  • Tencent subscription plans. Tencent Coding Plan (China mainland) and Token Plan (international) join the built-in providers alongside the pay-as-you-go TokenHub.
  • Codex: load saved cost totals in one pass per file instead of rescanning history for every day and model, preserving costs and token totals (follow-up to #3247). Thanks @robertoecf!
  • Menu bar: preserve measured card heights across cached provider tabs, preventing plugin cards from clipping on the first tab switch after launch.
  • Privacy: honor “Hide personal information” for project/source names and paths in cost history and project names in Usage & Spend, preserving costs, tokens, and stored data. Thanks @vinschger!
  • Automatic recovery reads project planning files only. Same-project session metadata and bounded replay now require explicit --metadata or --replay selection.
  • Cross-project session records remain quarantined, and replay output is bounded and nonce-framed as untrusted data.
  • Phase-status updates now use one shared shell and PowerShell directory lock and fail without mutation when the lock cannot be acquired.
  • 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).

Aug 29, 2026

Caveman
  • Hardened proxy redirect, authentication, cancellation, and TLS behavior.
  • Request-scoped MCP recovery and per-session native-runtime locking.
  • Safer Windows hooks and lifecycle handling.
  • Windows: .exe installer or portable. macOS: .dmg (Apple Silicon aarch64 or Intel x86_64). Linux: .AppImage (universal) or .deb (Debian/Ubuntu).
  • Debian 12+ ships without FUSE 2 by default, which AppImage requires. If ./omniget.AppImage fails with a libfuse error, run
  • bash sudo apt install libfuse2
  • Fix: graphify watch/update no longer silently drops a Markdown link to a semantic-backed document during a code-only rebuild — authored [[wikilink]] references are repointed onto the target file's…
  • Fix: T-SQL routines that parse to tree-sitter ERROR nodes — CREATE OR ALTER, the PROC shorthand, and bracket-delimited names like [dbo].[Get Widgets] — are now recovered by name over a…
  • Fix: graphify --help now lists the prs and provider commands and the export formats it actually supports (thanks @SyedFahad7).
Cumora
  • Mostly a hardening release for the engines that landed in 0.8.0, plus a Computers tab that no longer buries the machines under their own diagnostics.
  • Two follow-ups to 0.8.0's Gemini adapter. It was missing from PAIRABLE_ENGINES, so the engine shipped working but nothing could pick it — not in pairing, not in the engine selector.
  • The second is subtler and worth naming: triage set tools.core: [], which empties the built-in tool registry.
  • Start the preview findings service directly with codex-security serve [--port PORT], without Docker or an internal package path. The command reuses the existing service, state, and shutdown behavior.
  • Observe durable Deep Scan progress from the SDK with the optional onDeepProgress({ completed, active, maximum }) callback.
  • Make stopped-result handling explicit and stable. Read, list, and export operations no longer publish late retained results as a side effect.
  • Model selection: /model now handles the full selection flow, from quick choices to the complete catalog.
  • Fewer filesystem tools: fx now advertises eight fewer tools for better accuracy and more context, with terminal handling core filesystem operations.
  • Active-turn steering: While fx is working, Ctrl+Enter steers the active turn at the next model boundary. Enter still queues an ordinary follow-up, and late steering becomes the next queued turn.
  • @inspect variables: annotate a top-level const with a JSDoc @inspect tag and it becomes a live control in the stage inspector, below the background picker.
AI Memory
No release notes published.
CodeBurn
  • CodeBurn Desktop 0.9.23. Bundles the 0.9.23 CLI engine (matched pairing with the npm release, avoiding mixed-engine cache churn for users running both).
  • Manage automations across all local and remote hosts, with clearer destination-host information and a more polished table.
  • More resilient multi-agent orchestration, native chat, terminal sessions, and recovery across remote and SSH connections.
  • Preview remote HTML documents in Orca, alongside broad UI, workspace, mobile, and cross-platform reliability improvements.
  • Persistent agent memory. The agent can now remember what matters across sessions.
  • Provenance-aware figure workflows. The bundled scientific skills gain registered helpers for figure styling, multi-panel composition, and paper-ready narratives — built on immutable artifact…
  • Centralized credential management. GitHub tokens, connector keys, and connector sign-ins live in one place, with health status at a glance, guided recovery when a credential stops working, and…
  • Inline Mermaid diagrams and raster images now cycle only through visibly distinct sizes when clicked
  • Resized Mermaid diagrams invalidate cached message geometry and no longer leave blank placeholder space underneath
  • Clicking an inline image copies its pixels, while clicking a Mermaid diagram copies its editable source code
  • Add --mcp-config support and tool-call streaming to clai
  • Add @durable_operation for capabilities and a public backend API for third-party durable execution engines
  • Give instruction parts a stable InstructionPart.id

Aug 28, 2026

No release notes published.
  • Terminal capability overrides — Override detected terminal hyperlink, image, and truecolor support. See Capability Overrides.
  • Extension UI prompt events — Integrations can distinguish active agent work from time spent waiting for ctx.ui prompts. See Extension UI prompt events.
  • RPC queue clearing — Retrieve and clear queued steering and follow-up messages with clear_queue. See RPC clear_queue.
Omniget
  • Windows: .exe installer or portable. macOS: .dmg (Apple Silicon aarch64 or Intel x86_64). Linux: .AppImage (universal) or .deb (Debian/Ubuntu).
  • Debian 12+ ships without FUSE 2 by default, which AppImage requires. If ./omniget.AppImage fails with a libfuse error, run
  • bash sudo apt install libfuse2
  • Ported to MCP 2.0 — Every MCP surface in AG2, client and server, now targets the 2.0 protocol.
  • Server metadata — MCP servers can now expose metadata about themselves.
  • Conversation handling — Conversation handling updated for the MCP 2.0 model.
  • Native Antigravity 2.0: ECC installs to
  • Reliable uninstall: repeated selective installs keep a complete ownership
  • Safer migrations: OpenCode uses ~/.config/opencode, while OpenCode and
AI Memory
  • EMBEDDING_API_KEY, an optional embedding-only credential resolved ahead of
  • The standalone ai-memory-importer companion can now replay bounded generic
  • Added ai-memory workstreams, a read-only checkout-local list of recent
Claude Agent SDK
  • Updated bundled Claude CLI to version 2.1.251
Claude Code
  • Added PreModelSwitch and PostModelSwitch hook events (block, confirm, or annotate a model switch); SessionStart resume hooks now receive session staleness and the estimated re-cache cost
  • Added live streaming of a foreground subagent's tool calls and results to Remote Control clients (background subagents, the default, still show status only)
  • Added a Spend limit bar to /usage and a rate_limits.spend_limit status line field for developers behind a Claude apps gateway with spend limits
  • fix(chat-markdown): repair literal autolinks that swallow emphasis markers
  • fix: resolve spell check real-time highlighting and multi-language issues
  • fix(agent): polish subagent flow and running status
  • Antigravity: track token history directly from supported local SQLite conversations, without requiring a tokscale export.
  • Cursor: estimate API-rate costs when usage events omit prices, using cached or bundled pricing while keeping unknown costs unpriced and estimates separate from Cursor-metered charges.
  • Cursor: show empty cost-history windows without a decoding error, while preserving incomplete-history and malformed-response checks.
Kilo Code
  • #13512 13a9673 - Reference other Agent Manager worktrees with a searchable @ picker that prioritizes recently opened worktrees.
  • #13533 6d15d18 - Choose a separate model for conversation compaction in Context settings.
  • #13498 46bd29d - Review all committed and uncommitted Agent Manager worktree changes with /review worktree.
  • Durable Tasks and recovery
  • Resume interrupted Tasks after refreshing or restarting the app.
  • Preserve completed work without automatically repeating external actions.
  • GitLab sign-in, one more lock on agent context reads, and agents that can tell when a message was addressed to them.
  • Alongside Google, GitHub and Apple — including self-managed GitLab instances, so a team on their own install can use it as their identity provider.
  • A follow-on to 0.5.0's membership fix: the agent runtime's context read now enforces the tenant pinned in the JWT rather than deriving it from the request.
  • Release angular/v0.4.0
Stagehand
  • #2842 f86e609 Thanks @seanmcguire12! - update the MCP SDK to v1.30 and refresh supporting dependency versions
  • #2836 94a0ee4 Thanks @seanmcguire12! - fix cross-frame xpaths building logic for coordinates inside of iframes.
  • Store complete findings and embeddings in the preview findings service, with paginated listing, repository-scoped candidate retrieval, and durable duplicate groups.
  • Review potential duplicates from the CLI or SDK with dedupe --scan SCAN_ID --findings-url URL. Reviews run on the calling host and save accepted groups without replacing original findings.
  • Browse stored findings and duplicate groups in the service's read-only /dashboard, with search, repository filters, sorting, and record details.
Casdoor
  • migrate more code to web2 (bb9a309)
Langfuse
  • fix(ai)!: require LANGFUSE_AI_PROVIDER instead of defaulting to Bedrock
  • fix(billing): report Stripe webhook org lookup misses at the right severity
  • refactor(web): Clean up trace ui badge rendering
  • 跨语言 SDK 对齐:Python、Go、TypeScript SDK 补齐 find/search、context search、recall、资源、内容、Session、Skill、reindex 和管理接口;Python SDK 引入类型化 options,同时保留常用具名参数。
  • 检索可直接返回正文:find 和 list 模式的 search 新增 read_content,可把每个命中 URI 的可见全文放入 content 字段。CLI 对应 --read-content。
  • 支持 anydoc office converter
  • Durable notification inbox: runs publish authoritative outcomes and
  • Background subagents: a parent turn can spawn children that run and deliver
  • Persistent per-user sandbox containers on the local-Docker profile, reached
  • @mastra/playground-ui DataList API cleanup: removed variant="lined", flushLeft/flushRight, per-cell height, and DataList.MonoCell (use DataList.TextCell font="mono" instead).
  • Added a logger adapter contract (AdaptableLogger in @mastra/core/logger) for trace-correlated log output.
  • Update provider registry and model documentation with latest models and providers (7176362)
  • #3333 8f43674 Thanks @kimi-agent-bot! - web: Fix the command tool row rendering noticeably taller than other tool rows.
  • #3333 8f43674 Thanks @kimi-agent-bot! - web: Fix the first IME (or keyboard) character being silently swallowed after clicking the placeholder text in an empty composer.
  • #3333 8f43674 Thanks @kimi-agent-bot! - web: Fix switching the permission mode in one session changing it for every session; the permission mode is now scoped per session.
  • a92e8f6f7e92306ff37d5e93267bddb72f368fe2 MUL-6689: feat(agents): support MCP config for the Oh-My-Pi runtime
  • c0f5650e094d80a7ba321310c950f2ded0068cbb MUL-6735 fix(lark): recover from a rejected tenant_access_token instead of replaying it
  • 6df826fbee5d090205ee9cdea873d732ad83f9c3 MUL-6737: docs: correct the daemon poll interval default and describe the wake path
  • fix(dev): harden local cluster and gateway startup
  • fix(server): suppress expected session errors on shutdown
  • test(network): avoid DNS socket bind race
  • feat(plugin): add DeepSeek Harness memory adapter
  • fix(plugin): resolve DSH Viewer assets in clean builds
  • chore: sync MemOS Cloud OpenClaw plugin
  • Newly created templates and snapshots now record the versions of the components they depend on.
  • Architecture: fewer components to deploy — the network-agent component is merged into Cubelet, reducing inter-process RPC calls during sandbox creation.
  • Performance: optimized the eBPF network policy provisioning path, significantly reducing sandbox creation latency for sandboxes with network rules.
  • CodeBuddy agent framework. A fourth selectable agent framework joins Claude Code, OpenCode, and Codex — installed and managed from Settings with no separate login, running through the model…
  • Annotations. Select text in the transcript, tool activity, or file previews — or a point on an image — and send it to the agent as context.
  • Expanded OpenCode catalogs. OpenCode Go grows to 21 models and OpenCode Zen to 40, including the latest Claude, GPT, Grok, GLM, DeepSeek, Kimi, and Qwen families, with per-model endpoint…
  • feat(allowlist): add Handlebars and Mustache support
  • feat(config): validate provider existence before auxiliary fields
  • feat(agent): include churn stats (+N/-M) in the other-changed-files list
OpenCode
  • Fixed Azure authentication so Azure CLI sign-in works without requiring Bun.
  • New Azure DevOps connector for data sources
  • New You.com web search provider for chat and agent
  • New Serply web search provider for chat
  • fix(formatter): derive audio format from media type
  • fix(gemini): exclude tool-execution tokens from output_tokens
  • feat(formatter): use native multimodal tool outputs
  • Cline Desktop now ships on Windows: releases include a code-signed x64 installer, and installed apps auto-update on the same feed macOS does
  • Windows shell fixes: background processes (the sidecar, git) no longer pop visible console windows; updates now download in the background and install when you restart the app.
  • Tool results that return images — screenshots from browser or MCP tools — now render as inline images you can click to expand, with a carousel for stepping through multiple images, instead of raw…
  • agent-framework-core: Add configurable timeouts for waiting on the first background-agent task completion
  • agent-framework-core: Allow programmatic OpenTelemetry service metadata, resource attributes, and OTLP exporter configuration
  • agent-framework-foundry-hosting: Update Agent Server dependencies and expose parent FoundryToolbox constructor options
  • Route Heroku GLM models through zai_model_profile
  • Scope safe_download cookies to original hostnames
  • Fix DeferredToolRequests.remaining resolving cross-category result IDs
  • Fix: the incomplete-build shrink guard now stays armed when a chunk came back hollow, unparseable, or omitting files, so a run that silently lost content can no longer overwrite the existing graph…
  • Fix: graphify extract --force --code-only now fully rescans code (instead of skipping unchanged files and keeping stale import/alias resolution) while still carrying the existing document/semantic…
  • Fix: a hyperedge carried from a prior graph.json now has its members routed through the dedup survivor remap, so it no longer dangles when one of its members is merged away.

Aug 27, 2026

  • Connect a Discord bot from inside the product, live voice gets a faster engine, models fall over gracefully when one stumbles, and a cold boot makes 9 network requests instead of 339.
  • Live voice got a faster brain. Managed live voice now runs on Flux, which commits your turn the moment you finish speaking instead of waiting out a silence timer…
  • If a model stumbles, you won't notice. Managed inference profiles now fall back automatically when the primary model has a problem, so a bad model day stops turning into your error message.
Claude Agent SDK
  • Updated bundled Claude CLI to version 2.1.248
LangChain
  • Initial release
  • fix(langchain): name the content type MCP conversion could not handle release(langchain): 1.4.0a1 test(langchain): skip MCP tests on a pydantic older than mcp supports test(langchain): drive MCP…
  • Added --restricted (or CLAUDE_CODE_RESTRICTED=1): removes the built-in tools that run commands or code and WebFetch (unless named in --tools), keeps file tools inside the working directory…
  • Added experimental.cacheTtl ("5m" or "1h") to agent frontmatter: a per-agent prompt cache TTL used when no subagent TTL setting is configured
  • Added claude self-hosted-runner --client-label <label> (or SELF_HOSTED_RUNNER_CLIENT_LABEL) to override the label the runner registers with (default: hostname)
  • a2a: Port event action allowlist, agent card description source and resume flattening to v1 (1f898a6)
  • artifacts: Port app-scoped file artifact storage and layout-derived payload paths to v1 (0ae2bff)
  • artifacts: Port artifact reference scoping and path-segment validation to v1 (efc5bb8)
  • compress: session-aware /v1/compress (sidecar mode) + /v1/usage relay (4fa8802)
  • proxy: self-limiting session state for the compression-cache registry (826b600)
  • proxy: unify proxy and sidecar compression on one session engine (d12ea50)
Strands Agents
  • feat(ts): add model routing (f861b3392)
  • feat(python): add session_id property to Agent (1e5ff5866)
  • feat(python): inject an external cancellation signal into agent invocations (726b71a57)
  • TrustedRouter declarative provider
  • OpenCode Zen gateway as a declarative provider
  • Gondola as declarative OpenAI-compatible provider
  • feat(settings): select supported LLM providers
  • feat: add linux desktop installer build (CI + deb maintainer fix)
  • feat(automations): show a run's live phase
  • Promote conversational flows to stable
  • Record a created deployment with the given UUID
  • Enhance conversational flow documentation and APIs
  • feat(dev): unify local Kubernetes gateway workflow
  • docs(agents): clarify user-visible PR review feedback
  • feat(kubernetes): export driver traces over OTLP
  • Release v1.69.3
Cumora
  • Two more places the server could be pointed at an address it shouldn't reach, and a composer that stops eating your half-typed messages.
  • The Open Graph proxy validated the URL it was given, then handed it to fetch with redirect: 'follow'.
  • The same diff fixes a quieter one: the DNS check only looked at the first resolved address, so a hostname answering with one public and one private record passed.
  • [x] Native Foundry Local SDK on Windows (x64/ARM64) - the Foundry provider is now fully self-contained and no longer requires a separate Foundry Local install, with GPU/NPU-optimized model…
  • [x] GenieX runtime support for Windows ARM64 devices
  • [x] AWS Bedrock cross-region inference profile support - geo-prefixed profile IDs (us., eu., global.) now appear in the model dropdown and route correctly, plus manual region entry for regions…
  • A directory passed to pipecat eval run now reads .yml files as well as its .yaml ones, matching the scenario names a manifest resolves.
  • Fixed pipecat init repeatedly offering to build a Context Hub index that already exists, and the stale-index warning never appearing, on any index refreshed by Context Hub v0.5.3 or later.
  • Claude Code
  • Qwen Code
  • SWE-QA-Bench for repository-level, cross-file software-engineering questions
Casdoor
  • migrate the frontend from antd to shadcn/ui in web2 (a96940f)
  • 9aedd49 Thanks @omeraplak! - Add type-safe MCP annotations and metadata to tools and forward them through the MCP server adapter.
  • #1402 844939c Thanks @Iams4kura! - Preserve JavaScript replacement tokens in simple template variable values.
  • Fuzzy search in the MCP servers and connectors browser
  • Faster startup through additional lazy imports and deferred schema builds
  • Trust prompt is skipped when --worktree or --trust already grants access
  • Address devices by name and OS version — test runs now resolve devices by their human-readable name and OS version (e.g.
  • Mobile members work in local test runs — mobile participants are no longer limited to remote runs; they can join and execute in local test runs.
  • Evidence publishes on every mobile replay — mobile replay now emits evidence on each step, matching the behavior already in place for web runs.
  • Sheets: more reliable AI edits — sort, copy, and fill operate on raw cell values; large streamed workbooks now support formulas and sheet splitting.
  • Sheets: AI chat can export a sheet as a standalone file; faster conditional formatting on big workbooks.
  • Slides: scrollable notes pane at a readable height, plus text fidelity fixes for embedded fonts, gradients, and baselines.
  • Release Notes
  • Breaking Changes
  • @mehedi-107 made their first contribution in
  • Context Window Usage Indicator — A bundled plugin showing a composer usage ring, responsive breakdown popover with token counts, percentages, free-space reporting, and provider-reported…
  • Context Doctor — Automatically repairs malformed tool-call JSON before dispatch, handles incomplete or thoughts-only responses, supports XML suppression, and persists repaired output in WebUI logs.
  • Vision Sidecar Integration — Strict per-preset sidecar slot with Main-first routing, one-call multi-image analysis, parallel-worker image resolution, configurable timeout/max-token controls, and…
  • #2071 9620b58 Thanks @ben-reitz! - Make durable chat recovery unconditional for AIChatAgent and Think.
  • #2133 d536067 Thanks @mattzcarey! - Vendor the required PartyServer runtime into agents/lifecycle and add a reusable Durable Object lifecycle for startup, request interception, alarms, and…
  • #2058 381b9bb Thanks @ben-reitz! - Throttle chat UI updates by default in useAgentChat
  • <stage> / <scene> replace document roots — one scene per cut, camera state on the stage
  • New <adjustmentLayer>: a transform that rides above the clip below it, trimmed and slid in a row of its own
  • playbackRate, parent-relative timing, and containers that span their children
  • hermes update from an existing install
  • Fresh install: curl -fsSL | bash
  • commit: 57a266872237a4b79ace5d748181f4157965af8f
  • source tag: beta2026-08-27_19-51-33
  • regions: cn global
  • #3034 f0a6094 Thanks @sailist! - Add Remote Control as an experimental feature for accessing a local web session remotely.
  • #3157 491ebd0 Thanks @liruifengv! - web: Fix the slash-command and @-mention panels failing to open on mobile — both panels and the + menu are now grab-handle bottom sheets on small screens.
  • #3157 491ebd0 Thanks @liruifengv! - web: Add a flat/by-workspace tab to the mobile session list.
  • npm: bash npm install -g base44@latest
  • Homebrew: bash brew install base44/tap/base44
  • Version: 0.1.11 NPM Tag: latest
  • Restored Instagram login and fixed duplicate Facebook messages.
  • Improved WhatsApp account health and business profile visibility.
  • Improved Captain analytics, reasoning, and conversation context.
No release notes published.
Mem0
  • Usage telemetry: Anonymous usage events (deepseek.plugin.mounted, deepseek.tool.search_memory, deepseek.tool.add_memory) are batched to PostHog over native fetch and flushed in the background.
  • Request an advisory assessment of a completed patch with patch --assess-patch-risk. Add --create-pr to include its concise summary in the draft pull request.
  • Import GitHub code scanning alerts through the CLI or SDK for validation against a local checkout. Imports are read-only and preserve the upstream alert context. See GitHub alert imports.
  • Publish findings from CSV with publish scan --to cloud --csv PATH, or preview the upload without signing in or sending data with --dry-run. See Cloud publication.
  • feat(evals): track evaluator creation configuration
  • feat(web): allow serving build output from a dedicated asset host
  • feat(billing): implement CHB api client
  • add Amazon Bedrock Mantle (OpenAI-compatible) AI provider (bdbc3c7)
  • add LiteLLM backend (47db90f)
  • clamp max-concurrency in custom analysis to avoid deadlock and panic (6859e9c)
  • feat(session): register GroupingTask as a tracked TaskType
  • feat(rules): add Solidity and Vyper review rules
  • feat(action): opt-in completeness-gated checkpoint ranges
  • 修复了 Tutti Agent 认证初始化问题,提升 Agent 启动可靠性
  • 改进了 Agent 提供者配置流程,减少设置过程中的错误
  • 新增 Agent 就绪引导与故障排查文档,帮助快速定位配置问题
OpenWiki
  • #740 ec95f45 Thanks @colifran! - fix: finalize repository generation once when source changes during a run
  • Carry Bedrock guardrail trace in ModelResponse.provider_details
  • Honor provider_factory when cancelling Temporal responses
  • Cap dbos below 2.31.0 to avoid a missing opentelemetry-instrumentation-logging import
  • Added a /model <name> argument that switches to a model by name, slug or label and saves it as your default in one step, with the rest of the first matching name offered as ghost text while you type.
  • Improved the /effort hint so it completes what you have actually typed instead of always showing a fixed [low|medium|high] placeholder.
  • Improved artifact handling in conversations that produce many files by coalescing bursts of filesystem events into a single rescan.
  • Spanish interface. The full interface — onboarding, settings, conversation surfaces, native dialogs, and release notes — is now available in Spanish, joining the existing seven languages with a…
  • In-app source previews. Links in agent responses open in a sandboxed preview inside the app: hovering reveals the source title and full URL, and a click loads the page in the side panel with a…
  • Live notebook variables. A new Variables view inspects the running Python or R namespace — names, types, shapes, and previews — read-only, refreshed after each execution, with no kernel started…
  • fix(voice): agent and user state while a tool runs
  • chore: teach reviews to catch untagged PII logs
  • feat(agents): serialize the inline AgentTasks awaited from one turn
  • Add the JetBrains Agent Manager beta for creating, opening, organizing, renaming, deleting, and tracking worktree-based tasks and their sessions from the IDE.
  • Show Agent Manager worktree activity, changes, ahead/behind, pull request, failure, and attention badges with clearer row actions, menus, tooltips, and drag-and-drop reordering.
  • Add New, From PR, and From Branch tabs to the New Worktree dialog for clearer worktree creation and import flows.

Aug 26, 2026

  • Fixed Windows ARM64 launcher selection to prefer a native ARM64 executable when present and fall back to the published x64 executable through Windows emulation when it is not
  • Fixed stream URL tracking to emit active main-frame URL updates for full-document, History API, and fragment navigation, while rebinding correctly after active-tab changes and ignoring child-frame…
  • Fixed snapshot diff element references by resetting ref numbering for each diff, invalidating refs across URL navigations, and preserving the previous refs when a diff fails
  • Added image_url support for Gemini adapter, enabling external URLs via Part.from_uri().
  • Added Google Speech-to-Text v2 adaptation support to GoogleSTTService, so recognition can be biased toward domain terms using inline or referenced phrase sets.
  • Added MCPClient(tools_arguments=...), which injects extra arguments into every call of a tool.
No release notes published.
  • a2a: add native task mode support to RemoteA2aAgent (72f3ff5)
  • add ADK_MAX_LLM_CALLS environment variable to configure max LLM calls limit (75679db)
  • add create_data_agent tool to data_agent toolset (5cd38a0)
  • #724 57948ad Thanks @akyourowngames! - fix: keep hint/legend overlay inside the graph panel and make background clicks no longer clear the reader
  • #730 9f95289 Thanks @colifran! - fix: stabilize generated wiki formatting and claims hashes
  • #728 d219653 Thanks @colifran! - fix: prevent invalid optional OKF metadata from aborting wiki generation by repairing or removing it deterministically
  • Added the SendFeedback tool: when something goes wrong in a session, Claude can draft a feedback report for you to review and send from /feedback (turn off with the feedbackDrafts setting)
  • Added {id, text, cooldownSessions, priority} entries, tipsFile, and label to spinnerTipsOverride, so organizations can rotate their own tips alongside the built-in ones
  • Added a tip on Bash permission prompts pointing to auto mode, with a one-keystroke "Yes, and switch to auto mode" option
CopilotKit
  • Python SDK release: copilotkit 0.1.96
Feynman
  • Resuming a valid Pi session whose JSONL file lacks a trailing newline now repairs the append boundary before new research messages are persisted, preventing the next record from being fused into…
  • /btw and /btw:summarize now copy extension-registered providers, native providers, and temporary runtime API keys into their isolated child ModelRuntime.
  • Updated pi-web-access to 0.25.0. Researchers can route search and fetch calls through an explicit HTTP(S) proxy, retrieve bounded GitHub issue and pull-request documents with comments and review…
  • fix(email): stop habit-drip triggering on a per-workspace event
  • chore(desktop): bump version to 1.24.2
  • perf(auth): collapse the session lookup from seven round trips to one
  • A fifth BYOA engine, the dark theme finished off, and one more lock on attachment reads.
  • OpenCode joins Claude Code, Codex, Grok Build and Cursor Agent as a BYOA brain. Full wakes run opencode run --format json --auto, continue the emitted session id on the next wake, and record each…
  • Triage is deliberately not the same command: it runs through an injected cumora-triage agent whose wildcard tool permission is deny, and never receives --auto. The small brain judges; it cannot act.
  • Chat-On-Steroids-Setup-x64.exe — Windows x64 (Intel/AMD)
  • Chat-On-Steroids-Setup-arm64.exe — Windows on Arm
  • Chat-On-Steroids-macOS-x64.dmg — macOS Intel disk image
  • Qwen3.8-Flash
  • Qwen3.8-27B
  • Qwen3.8-2.4T-A95B
Zed
  • Added support for low reasoning effort for DeepSeek V4 Flash and V4 Pro. (#62577; thanks lingyaochu)
  • Improved JetBrains keymap behavior with CamelHump-style subword navigation for alt-left, alt-right, shift-alt-left, and shift-alt-right in editors. (#51540; thanks loadingalias)
  • Improved Markdown table column sizing based on content. (#61773; thanks timvermeulen)
  • LLM-generated session titles and terminal title
  • Queue selection and edit mode for queued messages
  • Git info is shown in the session header
  • feat(fetch): evaluate page expressions
  • feat(fetch): evaluate expressions from files
  • fix(csp): enforce frame-ancestors on child responses
  • Mastra Code LSP is now opt-in ("lsp": true); by default it won’t start language servers or offer lsp_inspect
  • @mastra/playground-ui: MarkdownRenderer streaming no longer paces word-by-word reveal (use useRevealedText), and CSS class mastra-markdown-arriving renamed to mastra-arriving
  • @mastra/playground-ui: SankeySignals now requires controlled selectedFrameId and onFrameIdChange props
  • feat(auth): read custom OIDC profile from the userinfo endpoint
  • fix(shared): truncate() no longer splits surrogate pairs
  • chore(storybook): Move non-component stories out of design-system dir
  • bf7627b2ffda9d597fde27eebc7e66122990e4b7 Merge pull request #627 from pinchtab/dependabot/npm_and_yarn/npm/js-yaml-4.3.1
  • d81a2a8b0ef9cd060b4cfdd5b315488f18306a18 Merge pull request #632 from pinchtab/fix/npm-global-install-home-divergence
  • d3ea1fecbfc12cc9d2f78562e23b00778da73057 Merge pull request #635 from valkyriweb/fix/daemon-owned-autostart
  • e6013e83e466af48afdd3d5c901071d191e01892 MUL-5928: fix(issues): bound board description previews
  • 8fda4f22d5429e8ece1fc56fdc7ef10cf4fd49d2 MUL-6016: feat(task): retry exact Codex capacity failures
  • 174c6525899ab43668ce2ead489d73b581643c74 MUL-6141: feat(runtimes): price Qwen/Kimi/Ark models and strip ':' provider prefixes
  • commit: c4a908856167b41078245e053ca2ac670f812685
  • source tag: beta2026-08-26_19-08-19
  • regions: cn global
  • Faster agent runs with many tools: Tool schemas are now derived once and cached across runs, cutting per-run overhead for agents that carry large toolkits.
  • Faster long conversations: Session history is loaded incrementally per turn, so response time stays flat as a conversation grows instead of scaling with its length.
  • PubMed request timeout: PubmedTools now accepts a timeout and passes it to both NCBI E-utilities requests, so a stalled PubMed response cannot hold a tool call indefinitely.
Cline
  • Fixed the background hub process ballooning in memory during long sessions — session status updates were broadcasting a full copy of the conversation transcript to every connected client, which on…
  • New files are now created with your platform's native line endings
  • Fixed the codebase search tool crashing on files that contain a single enormous line
OpenSandbox
  • OPENSANDBOX_EGRESS_SANDBOX_ID is now server-injected and rejected in request env — the server unconditionally sets the egress sidecar's sandbox-id attribution variable in both Docker and…
  • Sandbox lifecycle hooks (preStart / periodic), OSEP-0020 — CreateSandboxRequest.lifecycle is accepted, validated (reserved transports, duplicate periodic names, pool-allocation combinations), and…
  • preStart.timeoutSeconds limit raised from 300s to 3 hours — server and OpenAPI limits now allow up to 10800s for pre-start hooks; the periodic limit stays at 300s.
  • Sheets: CSV support — export the active sheet as CSV and save CSV files directly.
  • Sheets: closer Excel fidelity for charts, fills, pivot styles, pictures, and validation dropdowns.
  • Docs: DOCX fidelity fixes — image selection and anchoring, WordArt pictures, cleaner PDF export.
  • feat(rules): add Objective-C-specific review rules
  • feat(allowlist): add matlab support
  • feat(rules): add project review rule for providers.go
  • ROB-878 remove stale Sonnet 4.0/4.5 model recommendations from docs
  • Remove experimental Azure SQL toolset
  • ROB-893 Fix command injection vulnerability in Kubernetes toolset
BrowserSkill
  • chore(release): extension 0.1.7
  • fix: 退出后鼠标报告模式残留——dispose 窗口内自愈探针重写 ENABLE(#522)
  • Update README with core capabilities section
  • fix(test): timeline-rail 开场 settle 条件补齐顶部锚定不变量——修 #532 起的 CI 惯性红
  • Generated and editable session details. New sessions get an auto-generated title and description from the first message, and you can edit them at any time — Home cards now show what each session…
  • Per-call usage insights. When the framework reports enough data, each model call's tokens and context-window share are recorded, and the Context window dialog's Calls view becomes a per-call chart…
  • MCP client-config import/export. Import the standard mcpServers JSON used by other MCP hosts (multi-server files let you pick one), and export either an Open Science Connector or an MCP client…
  • Codex weekly resets recover cleanly: early backend resets refresh open usage cards without mixing accounts, spending reset credits, or losing confirmation across relaunches.
  • Codex spend stays visible during catch-up: established totals remain accurate while newer local session history finishes processing, without rebuilding existing caches.
  • Grok menus stay responsive: local session scans no longer block menu opening, and bearer billing restores real usage without losing account context.
  • Fixed syntax highlighting in the expanded python tool-call view: triple-quoted strings spanning multiple lines now keep their string color instead of only the first line.
  • Changed the default RLM maximum recursion depth for new sessions from 1 to 2.
  • Changed ACP prompt requests to resolve only after all causally admitted subagent and parent work has settled.
jcode
  • Desktop-owned sessions are marked as crashed when their client disappears unexpectedly, while deliberate detaches remain clean
  • Linux x86_64: available
  • Linux aarch64: available
  • Added /voice dictation, which transcribes speech straight into the prompt; press f5 or run /voice to start and stop it, and use the mic-serve subcommand to forward a local microphone to a CLI…
  • Added a cost field to the status line data model, exposing the unrounded estimated cost of the current session so a custom status line can display running token spend.
  • Improved the agent's code search by running an embedded ripgrep binary instead of shelling out to whatever the machine provides, so searches are faster and behave identically on systems with no…
  • Deprecate RunContext.capability_loaded and available_capability_ids in favor of capability_active and active_capability_ids
  • Make inclusive integer maximum reachable in TestModel
  • Reduce the default Temporal metric export frequency
  • Add machine-readable daemon health check
  • feat: allow disabling update checks
  • fix: redact CDP credentials from daemon logs
  • bb desktop 0.40.0
  • feat(desktop): add KLIPY GIF search to composers (8471049c430073474939336dfc6aa98272bc8762)
  • fix(desktop): respect automatic mention preference after send (e8cd7516e6df62c2a9025d7a821bf98e0e8f83b4)
  • fix(release): attribute desktop candidates to the operator (820a8589971df49bb9285a236ce1e2955a301abd)

Aug 25, 2026

  • Same-model research turns now preserve structured reasoning metadata through session persistence and model replay instead of flattening or dropping it.
  • When research history crosses a provider boundary, Feynman strips provider-private reasoning fields before forwarding it, preventing incompatible metadata from leaking into the next model's context.
  • Runtime preparation now repairs stale Pi AgentCore copies to the current maintained patch before verification, so restored and installed research runtimes cannot retain an older patch shape.
  • Updated bundled Claude CLI to version 2.1.246
Claude Code
  • Added a startup warning for Bash allow rules with a wildcard before the subcommand (e.g. Bash(git * main)), since they also match options inserted before the subcommand
  • Added an Auto mode tab to /permissions for viewing and editing auto mode classifier rules
  • Added the turn's completion time to the end-of-turn duration line, e.g. ✻ Sautéed for 23s · done 6:05 PM
  • Your assistant lands on your iPhone's Home Screen, Android gets a checkout, and a message you sent into Slack or Telegram is no longer set in stone.
  • iOS Home Screen widgets. Your assistant now lives between your weather and your calendar.
  • Android gets a checkout. If you're on Android, purchases now open a browser checkout (the same flow iOS uses), so upgrading doesn't require a detour to a laptop.
  • docs(nest): make commit attribution policy-neutral (22f32c99e2e983b7e015e8c058cac47438d05b2f)
  • fix(desktop-messages): preserve inline agent mentions with persistent addressing (b58de7cfa7a13b1a6dbb2fc269186439a6fb79a0)
  • feat(desktop): persist agent addressing across composer messages (7ba1197aa6b01616b5920487dd585a37d2e7f74d)
  • Chat-On-Steroids-Setup-x64.exe — Intel/AMD Windows PCs
  • Chat-On-Steroids-Setup-arm64.exe — Windows on Arm
  • Chat-On-Steroids-Extension.zip — standalone companion extension
  • feat(cli): add disable-js fetch mode
  • fix(renderer): block javascript URLs when scripting is disabled
  • fix(renderer): fix missing parser continuation
Paperclip
  • Accepting the background-service prompt during an npx onboard no longer installs a broken service.
  • Onboarding no longer offers a foreground start the service already covers. After a successful service install, interactive onboarding still asked "Start Paperclip now?" — accepting ran a second…
  • Onboarding ends at the dashboard instead of a dead stop. After the service starts, onboarding now waits for it to report the endpoint it actually bound…
  • Fixed <kbd>ctrl+a</kbd> after v0.91.0.
  • Improved copying by selection with wrapped lines
  • Added documentation on how to set a custom Catwalk URL
  • 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 now…
  • Flash engine: the local default (mode="standard" keeps the classic LLM pipeline).
  • 🎛️ Take full control of native harnesses: switch Claude Code permission modes live (Manual, Auto, Accept edits, Plan) with shift+tab or from settings, /rename syncs back to the UI, model pickers…
  • ⏱️ Automations get real guardrails: scheduled tasks can cap LLM spend per firing with max_cost_usd and set a Claude Code permission mode (e.g. Accept edits, Bypass permissions) per task.
  • ⌨️ A faster web and desktop flow: Cmd/Ctrl+N starts a new session anywhere (including the macOS app), ⌘K and session-switch hotkeys stay live in the terminal and composer, the task tracker moves…
  • Release v1.69.1
Paseo
  • Command Center now matches partial multi-word queries in any order and ranks visible results by relevance
  • Fixed upgrades placing persisted Side panel tabs inside Explorer instead of an ordinary side pane
  • Fixed Escape leaving Settings open on desktop and web
CodeBurn
  • CodeBurn Desktop 0.9.22. Bundles the 0.9.22 CLI engine (matched pairing with the npm release, avoiding mixed-engine cache churn for users running both).
  • Fix: Ruby methods whose names end in !, ?, or = now keep distinct node ids, so save and save! (or foo and foo=) no longer collide into one node.
  • Fix: a Ruby call on a qualified constant receiver (ActiveRecord::Base.transaction) now matches the receiver's full constant path, so it no longer binds to an unrelated lone class named Base.
  • Fix: a CommonJS member export wrapped in a higher-order function (exports.x = wrap(fn), module.exports.y = onCall({...}, handler)) is now captured, reaching through the wrapper to the function it…
  • add enableGroups switch to the LDAP edit page (5f08e4c)
  • Full Changelog
  • #581 fab0a3f Thanks @JHSeo-git! - feat: adopt okf v0.2 with code-owned generated provenance
  • #638 c1ca21b Thanks @colifran! - feat: add grounded claims for self-correcting code wikis
  • #713 4882ba3 Thanks @colifran! - feat: replace repository generation with a resumable page-job lifecycle
No release notes published.
  • feat(server): add an opt-in bypass for the multi-worker in-memory queue guard
  • fix(ci): keep skipped package releases green and stop sending target_commitish (release-1.11.4)
  • fix(security): backport the 1.12 PVR security fixes to 1.11.5
  • feat(docker): export driver traces over OTLP
  • fix(proxy): normalize trailing-dot CONNECT hosts before policy evaluation
  • feat(sandbox,podman): trust corporate CA for https:// proxies and intercepted TLS
  • fix(tmux): route subagent panes to attached parent session
  • fix(smartfetch): set parentID on secondary sessions to suppress TUI notifications
  • fix(task-result): return status for running background tasks
  • Slides: new Effects pane (shadow, reflection, glow, soft edges) and true vertical text.
  • Slides: copy and paste elements and images across slides, decks, and windows.
  • Slides: one-click font install with missing-font detection; AI-generated text boxes now size to fit their content.
  • Dark mode, Windows agents that survive closing the terminal, and a server-side fetch that could be pointed anywhere.
  • A chat attachment carried a URL, and the agent wake fetched it. Since the accompanying storage key was optional, any conversation member could omit it, supply an arbitrary URL, and have the server…
  • Attachment content now resolves only through a server-generated storage key. The write path refuses to persist an attachment that doesn't map to Cumora storage, and the read path regenerates the…
  • Custom OpenAI-compatible providers can configure reasoning capability and default effort per model
  • Custom gateways can disable model-name reasoning heuristics that caused unsupported reasoning_effort parameters to be sent
  • Linux x86_64: available
  • E2B sandbox backend: agent shell/code execution can now run on E2B cloud sandboxes in addition to the local sandbox. Select the backend with DIFY_AGENT_RUNTIME_BACKEND.
  • Build-time Home Snapshots: when an agent build is applied (published), the agent's sandbox home directory — installed packages, prepared files, working state — is captured and subsequent runs of…
  • Workspace-level Skill management: Skills are reusable, versioned capabilities packaged with code and tool definitions that agents can discover and invoke.
  • Stop whole-schema claim test fixtures deleting other xdist workers' operations
  • docs: changelog and blog post for v0.9.1
  • docs: rework the topbar, page chrome and skill installer
  • Client-pool reconcile interval is no longer configurable — SandboxPool.Builder.reconcileInterval(...) and the corresponding PoolConfig setting have been removed.
  • Scalable asynchronous client-pool warmup — Pool refill now separates create, readiness, prepare, post-prepare health checking, renewal, and commit into a staged asynchronous pipeline.
  • Pool warmup tracing and operational summaries — Applications can opt in with ConnectionConfig.enableTracing(true) to emit OpenTelemetry traces covering queue wait and every warmup phase, using the…
  • Reverts #14889 ( released on 0.88.3 ) as it introduced a CPU spike in worker containers for flows that use old versioned pieces with large dependencies like google suite pieces .
  • 39ddc82c9dc25494286db4fbc07382d209527f86 MUL-6228: resolve full-UUID issue refs locally instead of a resolver GET
  • 8c563b49bede267267ae81d7988099e051f5c049 MUL-6555 feat(issues): preserve source context for sub-issues
  • 11ab949dc97fb12b653ef19f2619a09a6eab91c3 MUL-6611: docs(self-host): route /health through the proxy, add single-origin example and release-tag note
  • commit: 1e165ea83c806a62439fe6b7ad3ac2279f9f3d87
  • source tag: beta2026-08-25_17-57-50
  • regions: cn global
  • Plugins, MCP, Skills, Rules, Hooks, and Tools are now one Customize hub with tabbed sections and live counts.
  • Redesigned the Models page: providers are grouped into Connected, Popular, and All with their auth kind and configuration status instead of per-row toggles.
  • Voice input moved to its own Settings → Voice page that only offers connected transcription-capable providers and preselects a default model.
  • 新增 OpenCode 客户端接入
  • 新增 DeepSeek Harness (dsh) 接入 —— DeepSeek 官方 agent harness 的
  • 新增 Codex CLI 接入
  • Fixed Cloudflare AI Gateway routing for third-party providers so non-Workers models work through the gateway's REST API. (@superhighfives)
  • Fixed Anthropic models through Cloudflare AI Gateway by converting dotted model IDs like claude-haiku-4.5 to the dashed slug Anthropic expects. (@superhighfives)
  • Fixed parent session IDs being sent in request headers for session-aware providers.
Codex Security
  • Bug fixes and reliability improvements for cloud publication, including access checks, recovery handling, and skipping findings that were already recorded.
  • feat(publish): check Linear access and skip recorded issues
  • fix: reconcile Linear publication recovery identities
  • feat(agent): group semantically related files for multi-file review
  • feat(cli): add --output flag to write review/scan results to a file
  • fix(diff): surface git's own message when a diff command fails
  • Added private proxy CA trust for locally launched Chromium on Linux. Use --ca-cert <path>, AGENT_BROWSER_CA_CERT, or caCert in config and MCP to import a PEM bundle or DER certificate into an…
  • Added a bundled protected Vercel deployments skill that guides agents through short-lived Trusted Sources OIDC authentication, authorized automation bypasses, and explicit human handoffs for…
  • Settings → Appearance now lets you choose the typeface across four regions — app, chat, markdown and code — from a searchable list of fonts installed on your machine, plus a matching font-weight…
  • The archived conversations list now pages in the rest of each group instead of silently truncating it — a "load more" button reveals older archived chats a page at a time.
  • The agent's plan / to-do snapshot moves out of the message stream into a collapsible bar pinned above the send box, and a bug that could duplicate the plan card while destroying a neighboring…
dsh-TUI
  • <!-- Release notes generated using configuration in .github/release.yml at v0.9.2 -->
  • Full Changelog
  • Added skill icon and visual branding support across the CLI, displaying emoji icons declared under metadata.icon in SKILL.md frontmatter across the /skills catalog list view, detail inspection…
  • Improved @ file path autocompletion by indexing empty directories alongside files in ripgrep search results, allowing unpopulated and directory-only workspace structures to be discovered and…
  • Improved permission management by automatically granting workspace-scoped read access under the default review mode, eliminating repetitive approval prompts for reading or listing files within the…
  • Session references (#). Reference another session directly in the composer. The inserted chip navigates back to the target session, and the agent receives read-only access to the referenced…
  • Side-chat advisories in running main turns. Advice from a side conversation now reaches the main agent while its turn is still running, through each framework's native follow-up channel instead of…
  • Long pasted text as attachments. Plain-text pastes over 10,000 characters or 300 lines become a managed attachment with a card in the composer.
  • Changes since langchain==1.3.16
  • release(langchain): 1.3.17 fix(langchain): frame custom HITL rejection reasons chore(deps): bump minor and patch dependencies
  • Add a LangChain migration skill
  • Add GLM-5.3 support to ZaiModel
  • Fix TestModel generation for equal inclusive bounds
  • Terminal presentation: /appearance, /input, and /maxxing have been removed along with their saved settings. fx now uses the same input and transcript layout everywhere.
  • Foreground command timeouts: terminal.exec calls now require timeout_ms between 1 millisecond and 10 minutes. Use terminal.start for services, watchers, GUI apps, and other long-running work.
  • Remote MCP servers: /mcp add --transport http <name> <url> now saves or replaces a remote Streamable HTTP server and reloads MCP immediately. The existing local stdio form is unchanged.

Aug 24, 2026

  • Paste a Linear URL and kane-cli does the rest — Linear issues and documents are now recognized sources.
  • Workspace login pages are rejected clearly — if a Linear URL points to a gated workspace page that requires a login, kane-cli refuses loudly with a plain error message instead of silently scraping…
  • Attachments and images are extracted from issue bodies — inline images and linked media in Linear issues are pulled out and made available to the run, not silently dropped.
Cline
  • The first-launch "Try ClinePass" dialog no longer advertises the $4.99 first-month promo, which is ending
  • The hub's event log is now capped at 64 MiB on disk. Events carrying full session snapshots could previously grow the log to tens of gigabytes on a long-running hub, since deleting rows never…
  • Refreshed the model catalog. Adds two providers (AgentRouter and Opper) and updates model lists and pricing across providers.
  • . fix(rate-limiter): promote parked jobs when a concurrency slot frees ( #15033 ) @MrChaker
  • . this release has been ported only on top of 0.87.0 as hotfix and it's currently planned to be ported to 0.89.0
  • > 📌 Upgrading a self-hosted instance? Review the Breaking Changes page for any actions required before you upgrade.
  • SKY-15270: Show Copilot only the arguments each browser tool accepts (9693986ec)
  • Add the selector engines Skyvern's producers emit to skycdp (SKY-14071)
  • Capture post-run page evidence from the run's own browser session (SKY-14254)
Mem0
  • Automatic recall and injection: Mem0MemoryStore.search() runs every turn through the MemoryManager, so relevant memories are searched and prepended to the prompt with no explicit tool call required.
  • Server-side extraction: add_messages() renders raw conversation turns to text and hands them to Mem0's own extraction pipeline (infer=True), so enabling extraction skips an extra client-side model…
  • Verbatim writes: add() stores a single fact exactly as given (infer=False), the sink used by the add_memory tool or a client-side extractor.
No release notes published.
CodeBurn
  • Download the .msi below and run it. The tray app reads everything through the CodeBurn CLI, so install that first
  • npm install -g codeburn
  • Requires codeburn 0.9.9 or newer and the WebView2 Runtime (preinstalled on Windows 11 and recent Windows 10 updates; installed on demand otherwise).
  • fix(renderer): retire failed committed navigations
  • refactor(renderer): model committed navigation state
  • fix(renderer): resume stylesheet lifecycle after document.write()
Ziva
  • Godot 4.2+
  • Active Ziva account
  • Fixed a potential filesystem sandbox escape when running extensions.
  • Fixed GitHub Copilot Chat authentication and API routing for GitHub Enterprise Cloud
  • Fixed project-level language server settings being ignored by extensions built with extension API versions before and including v0.1.0.
  • Chat-On-Steroids-Setup-x64.exe — Intel/AMD
  • Chat-On-Steroids-Setup-arm64.exe — Arm (Snapdragon, Surface Pro X)
  • Chat-On-Steroids-Extension.zip — the extension on its own, for loading unpacked
  • Add an Agent.clone() method that returns a new Agent with the same configuration, optionally replacing some init parameters: variant = agent.clone(system_prompt="Answer in German.").
  • Added a link_format parameter to both PyPDFToDocument and PDFMinerToDocument components, matching the existing functionality in DOCXToDocument.
  • Agent now returns an exit_reason output reporting why the run stopped, making it easier to route the Agent's output downstream (for example with a ConditionalRouter).
  • Removed outdated OpenCode Go first-month discount messaging and pricing.
  • Fixed OpenCode device login links when servers return relative verification URLs or use a base path.
  • Fixed textVerbosity being sent to OpenAI-compatible providers that do not support it. (@joelstucki-taulia)
No release notes published.
  • Search scanned PDFs with private, on-device OCR — no uploads required.
  • Ask AI directly from selected cells and slide objects.
  • Work faster with Word-style shortcuts in Docs and active row and column highlighting in Sheets.
  • Feature: graphify merge-graphs now links a type declaration that two repos share — same fully-qualified namespace and name, from different repos — with a same_type_as edge, so a shared contract…
  • Feature: C# properties now emit a member node per property, so a property is visible as a class member alongside fields and methods.
  • Fix: when a C/C++/Objective-C declaration and its definition merge into one node, the node now points at the definition site (the implementation body) instead of the header prototype.
  • Tool result offloading: Agent(offload_tool_results=True) / Team(...) writes any tool result over 16,000 chars to AgentFS and leaves a short envelope (preview, size, result_id) in the message.
  • Media offloading: media_storage=S3MediaStorage(bucket=...) on an Agent/Team/Workflow uploads images, audio, video and files to local disk, S3 or GCS before persistence.
  • CodeMode: CodeMode(tools=[...]) swaps a wide tool schema for one programmable IPython kernel that persists across a session — the model writes Python and calls tools as awaitable handles…
  • fix(gemini): remove unsupported $schema from tool parameters
  • feat(workspace): share workspace across different agents
  • feat(workspace): isolate skills per agent, select them per session
  • Fixed workspaces crashing on Android tablets and foldables in non-compact layouts
  • Fixed app settings resetting when builds sharing a profile recognize different settings
  • fix(skills): support valid GitHub SKILL.md links
  • feat(provider-registry): remote catalog hot-update without an app release
  • refactor(agent-runtime): unify declarative tool approval policy
No release notes published.
  • [BREAKING] A new Slack experience — per-agent provisioned Slack apps, agent spawning from Slack, and UX improvements — is available to classic single-bot Slack installs.
  • /add-codex now pins @openai/codex 0.146.0. The previous pin (0.138.0) defaults to GPT-5.4, which OpenAI retires from Codex on 2026-08-31 — codex-provider agents ride the CLI default model, so…
  • [BREAKING] Agent mailbox access now goes through storage-neutral host and runner registries.
  • PowerShell tool — Use optional native PowerShell command execution on Windows. See PowerShell Tool.
  • Safer managed updates — Stage, verify, and atomically activate updates for installer-managed installations. See Install and Manage.
  • Model and thinking controls — Select thinking levels with /thinking, search defaults, keep selections session-scoped, and persist them explicitly with Ctrl+S. See Models and Thinking.
Cumora
  • A Sign in with Apple hardening fix, and Windows BYOA agents working end to end.
  • The native Apple sign-in endpoint accepted an email in the request body and used it when Apple's signed token didn't carry one.
  • Now only two sources can answer that question: an already-linked Apple sub resolves from the email persisted against it, and an unlinked one requires an email_verified claim in Apple's signed token.
  • bdcb231ba00415c952e0d46ca6a90ae9273b4ff3 MUL-5466 fix(agent): deliver the qwen prompt on stdin, not argv
  • 8f9c766e40d950565efdbc978c512b424e2a8d0d MUL-6460: feat(brief): inject the workspace status catalog into the agent brief
  • 668a4e06c32799c66460032b75c68a0920d20322 MUL-6464 fix(agent): support ACP terminals for Kimi runtime
  • None called out in this changelog.
  • Added graceful shutdown options for generated servers. Configure how long in-flight HTTP requests may drain, or disable Mastra's built-in signal handlers when managing the server lifecycle yourself.
  • The session now marks a message you send while the agent is working, and every client gets it for free.
  • Codex inside ChatGPT.app detected: Adaptive (agent-aware) refresh recognizes the ChatGPT-bundled Codex and holds the 5-minute active cadence while it works.
  • Antigravity without the desktop app: menu-bar quota refresh reuses an already signed-in agy CLI, and logged-out states show sign-in guidance instead of "Launch Antigravity".
  • Faster OpenCodex spend: pricing each usage entry once drops the spend refresh from ~12 s to ~3 s on a 35k-entry log.
  • 修复嵌入检查:解决了嵌入审查发现的问题,提升了插件的稳定性。
  • 增强输入处理:加强了对超大嵌入输入的处理,避免了潜在的崩溃问题。
  • 容错处理:修复了对缺失开发者对等冲突的容忍性问题,提升了系统的健壮性。
  • Integration discovery can now investigate a candidate's capabilities, compatibility, setup, pricing, and limitations before selecting it
  • Subagent models can now be chosen from the full model picker and stay synchronized across remote sessions
  • Model thinking is now shown by default
  • fix: restore main, add CI, and land 97 queued PRs
  • fix(readme): restore broken star history chart
  • docs: document scholarly retriever configuration
  • Windows x64:下载 GenericAgent-Desktop-Windows-Portable.zip,解压后双击 GenericAgent.exe。卸载时运行 uninstall.bat,再删除解压目录。
  • Linux x86_64:下载并解压 GenericAgent-Desktop-Linux-Portable.tar.gz,执行 chmod +x GenericAgent.AppImage 后运行它。卸载时运行 uninstall.sh,再删除解压目录。
  • macOS Apple silicon (arm64):打开 GenericAgent-Desktop-macOS-aarch64.dmg,把 GenericAgent.app 拖入 Applications。
  • commit: 4537b5e4adf0a0e5e8d5ea3b18ddce41298dfc6a
  • source tag: release2026-08-24_12-08-08
  • regions: cn global
  • xAI (Grok) OAuth subscription. One subscription account works across all three agent protocols — Claude Code (Anthropic Messages), OpenCode (Chat Completions), and Codex (Responses) — through the…
  • Governed Marketplace Specialists. Marketplace-installed packages carry an explicit marketplace origin: publisher content is read-only (icon, color, and enabled remain editable), manual ZIP…
  • Notebook cross-run dependency tracking. Completed Python and R runs are analyzed in-process with tree-sitter WASM — aliases, root-object mutations, classes, S4/R6 objects, copy/reference…

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.