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: Jul 27 – Aug 2, 2026

201 stable releases from 79 tracked agents and frameworks in the week of Jul 27, 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 2, 2026

  • refactor(mcp): rename packages/mcp-v2 to packages/mcp and de-version all user-visible surfaces
  • fix(desktop): truncate long base-branch name in 'from' label
  • feat(plugin): add Superset orchestration skill
  • The model selector now supports search, so you can find the model you need faster.
  • Batch element selection is now supported.
  • Conversation navigation has been streamlined.
  • Memory controls add opt-in retrieval caching, reranking, typed-fact extraction, migration-aware configuration checks, content scanning, and auditable recall paths.
  • SOP execution gains per-SOP admission policies, quorum-based approval brokering, checkpoint editing, centralized fan-in adapters, and stronger deterministic gate handling.
  • Operator surfaces add dashboard-driven upgrades, channel relinking, agent renaming, runtime-context display, searchable keybinding help, and a more reliable terminal experience.
Casdoor
  • keep password requirement popover open until requirements are met (e7884bb)
  • Fixed Option+Left stopping to the left of punctuation between words Video URL
  • Fixed a panic when selections are added by tab-expanded column
  • Fixed Gruvbox parameter colors
  • Immediate specialist handoff. Switching specialists mid-conversation now hands the in-flight turn to the new specialist right away instead of waiting for the next turn boundary, so context is not…
  • Completed-turn runtime icons. Each completed agent message now identifies the ACP framework and model that produced it, with framework and provider icons alongside usage in the token popover and…
  • Context-usage persistence. The context-usage indicator now survives application restarts.
  • feat(manifest): run manifest coverage contract for review
  • fix(telemetry): honour a base path in the OTLP HTTP endpoint
  • fix(vscode): use shell mode for spawn on Windows to detect npm correctly
  • Behavior-guard tests: log-fetch transparency eval, fix permissions eval setup
  • Fix operator image startup by removing holmes package dependency
  • ROB-679 Log the TOOL_MEMORY_LIMIT_MB remedy when a tool is OOM killed
  • macOS (Apple Silicon, signed + notarized): GenOffice-0.4.110-arm64.dmg
  • Windows (x64, NSIS installer): GenOfficeSetup-v0.4.110.exe
  • smithers-orchestrator → smthrs (and every subpath: smthrs/ui,
  • @smithers-orchestrator/* → @smthrs/*
  • /** @jsxImportSource smithers-orchestrator */ → /** @jsxImportSource smthrs */
  • Self-update is now seamless: a live progress bar during download and a graceful in-place reload when it finishes
  • Todos you are working on stay pinned in a band at the top of the viewport while you scroll
  • Ctrl+L is a true terminal-style clear: the screen blanks, history stays above, and the prompt sits at the top
  • Added input priority and per-client delivery settings to the stream server: each WebSocket connection now splits into a reader task and a writer loop, so clicks and keystrokes dispatch to the…
  • Added configurable stream encoding: frames were hardcoded to jpeg quality 80 at the session viewport, so frame rate was the only bandwidth lever.
  • Frame delivery is now latest-wins. The stream server holds only the newest frame and reads it at send time, so frames produced while an earlier one is still being written are skipped.

Aug 1, 2026

  • [BREAKING] iMessage unified into one imessage channel with two backends via /add-imessage: Local (this Mac's chat.db via the Chat SDK) or Hosted (native Photon via spectrum-ts, no Mac relay).
  • [BREAKING] Provider-agnostic memory. All providers now share one OKF v0.1-compatible memory/ tree, while persona lives in instructions.prepend.md.
  • New groups can inherit an instance-wide default provider. DEFAULT_AGENT_PROVIDER sets the provider used when a new agent group is created without an explicit provider.
  • PWF_PLAN_ROOT binds resolution to a project root by absolute path. PLAN_ID is a cwd relative slug and structurally could not name a nested project's plan at all.
  • An ambiguous cwd refuses to inject. When the plan came from the shared pointer or the newest by mtime fallback and a project one directory below has its own live plan, nothing is injected and the…
  • Applied to the canonical dispatcher, the .codex/hooks.json route, the Cursor hooks, and the shared resolver in both shell and PowerShell.
  • forward the request language to the Custom HTTP Email provider (2840e2a)
  • Added token usage metrics to AWSNovaSonicLLMService, which now emits LLMUsageMetricsData from Nova Sonic's usageEvent.
  • Added an http_client parameter to OpenAITTSService and Whisper-based STT services (BaseWhisperSTTService, OpenAISTTService, GroqSTTService), so a custom httpx.AsyncClient — e.g.
  • Added the ability for users to provide a local image to the LemonSlice transport to be used as the avatar image.
  • Stop button and /stop API — Cancel a running agent context without deleting it. The chat bar now shows a stop button while the agent is working, and a new /stop endpoint is available for…
  • UI asset bundler and service worker — A new build-time bundler discovers same-origin HTML, CSS, and JS assets and delivers them as a versioned JSON bundle.
  • WebUI extension manifest — The bundler and server now support an extension manifest, allowing enabled extensions to inject their own entry files into the bundle.
No release notes published.
  • Mobile remote access. The same localhost web UI can now be reached from a phone or tablet through Remote.It pairing.
  • Conversational specialist customization. Specialist agents can now be created, scoped, switched, and deleted entirely in conversation through the agent, powered by a new host.agents control-plane…
  • Message timing metadata. Each user message now shows a stable sent timestamp and each completed agent turn shows a finished timestamp, elapsed runtime, and a usage popover with input, cache, and…
Mem0
  • Dependencies: Patched high and medium severity dependency vulnerabilities via pnpm.overrides (brace-expansion, js-yaml)
  • Stopped MCP SSE connections from getting stuck in reconnect loops after server error responses.
  • Fixed provider model configs that use interleaved reasoning fields like reasoning_text or custom field names.
  • Open external links in the system browser instead of inside the app.
  • feat: support the /improve command for the local git provider
  • feat: optional provider-side request attribution via the OpenAI user field
  • fix(litellm): surface the real provider error instead of RetryError
  • A non-loopback bind adds --mint-token automatically, and a wildcard bind
  • Review printed URLs before sharing them: they carry a login token.
  • When no gateway is reachable, the CLI error now names the runtime state
  • fix(llm): drop extra_body.stream from non-streaming requests
  • fix(diff): honor .gitignore negation patterns
  • fix(action): group 422 fallback inline comments into a single review
  • Allow MCPToolset clients to skip optional MCP tasks via prefer_tasks
  • Default Gemini VALIDATED tool mode on supported models
  • Stop advertising tool search for capability-owned deferred tools on Anthropic
  • feat: expose original callable through wrapped
  • fix: report input guardrail results when a tripwire aborts the run
  • fix: consolidate same-exception re-raises into bare raise across all modules
  • Your assistant is easier to reach on iPhone. Talk to it from the lock screen, while using another app, or with Siri and the Action Button.
  • Keep your conversations organized. Create groups for work, projects, trips, or anything else. Drag chats into a group, give it an icon, and arrange your conversation list the way you want.
  • Connect plugins more safely. Plugins can now receive information from other services after you approve the connection in settings. You stay in control of what gets connected.

Jul 31, 2026

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
  • Improved search relevance so queries return more useful results.
  • More robust ingestion: fewer failed uploads and clearer error messages.
  • Stability fixes that prevent rare crashes and reduce memory usage.
  • add parent terminal grouping and TTL pruning to ADK CLI telemetry (99bbd83)
  • capture TTY connectivity in CLI environment telemetry (c12a025)
  • cli: implement early telemetry recording for long-running web servers and log successful exit code upon routine teardown (77726c5)
  • 9e3ed61cfb6e1a6f087fef45e3a76faa9289a01e: feat(bedrock): retry the turn automatically after AWS SSO re-auth (@jan-xyz)
  • 7810d0382d8d7699ddbaa2f9bbc484d5e6daaace: feat(config): apply top-level env vars on startup (@jan-xyz)
  • 155072b32603bb0ebc2594fc90c2e47e25d2f566: feat(noninteractive): make the spinner text use terminal default (@taciturnaxolotl)
  • Revert "chore(release): release Buzz Desktop version 0.5.3" (bb34bc4d98fe4dabe847046103ac5e2859917ac5)
  • chore(release): release Buzz Desktop version 0.5.3 (d12b3d6a79d56a95fc99ce4fadd2d2235d5a3131)
  • feat(desktop): import local Pocket voices (c104eecfb38620de2c35c7e20a716f8658b5a6b1)
  • #514 b8c510f Thanks @colifran! - chore: setup changeset tooling for automated releases
  • #530 1695c3f Thanks @Monkey-wusky! - fix: allow comma in model id for gateway/proxy routing identifiers
  • #533 fdfdfd8 Thanks @jyje! - fix: keep release workflow opt-in on forks
QM
  • Restrict the web-ui model picker to the org allowed-models list
  • Use @latest in the qm init bootstrap instead of a version placeholder
  • fix(server): bind gateway listeners before sandbox resume
  • fix(proxy): retry with backoff on transient accept errors instead of exiting
  • fix(tasks): scope pre-commit to lint checks
Codex Security
  • feat: Automatically match scan comparisons
  • feat: automate verified npm and GitHub releases
  • fix: read container manifest digests without broken pipes
  • Clicking a filename search result now clearly highlights the file in the Explorer tree with a full-row selection and scrolls it into view, so you can see where the hit actually lives.
  • Tab now accepts the highlighted @-mention item, just like Enter — a small but welcome shortcut when referencing files in chat.
  • Token usage is now reported for the direct-CLI backends (Claude and Codex), so the context usage indicator works on those conversations too.
  • For the installer path, run CodeWhaleSetup.exe; it installs codewhale.exe, codew.exe, and codewhale-tui.exe under %LOCALAPPDATA%\Programs\CodeWhale\bin and adds that directory to the current-user…
  • Extract the archive for your machine: codewhale-windows-x64.zip or
  • Run install.bat (copies to %USERPROFILE%\bin)
  • Collect skill usage events
  • Remove migrated AMP documentation
  • Update security reporting guidelines
  • Personal specialist agents. Define reusable agent profiles with their own name, avatar, persona, and scoped capabilities — a full-access catalog or a symmetric skills/connectors whitelist — and…
  • Scoped permission management. Tool-call approvals are no longer tied to a single conversation or an ACP-specific lifetime.
  • Export conversations as Markdown and PDF. A whole conversation can now be exported as a portable Markdown or PDF document through a normalized public export model (not the internal session JSON)…
  • Sub-flows now run back-to-back automatically — every recorded sub-flow executes in a single spawn, so a multi-step plan runs end-to-end without manual re-triggering.
  • Negative-verdict verification — failing checkpoints are now verified against stored memory before surfacing, reducing false failures.
  • Upload and analyze Word documents end-to-end — .docx files are unpacked, their text projected deterministically, and images inventoried and delivered alongside PDFs and other formats.
  • Fixed conflicts that could occur when installing the same provider multiple times
  • Improved install flow to prevent duplicate tasks from interfering with normal installation
  • Internal packaging and update validation improvements
  • ChannelHandler context parameter is now required: the 4th parameter is ctx: ChannelHandlerContext (non-optional).
  • Added support for Code Mode transports that provide their own execution boundary.
  • Added retry callbacks for stream error policies and ensured explicit matcher policies override provider retry metadata.
  • Release Notes
  • Breaking Changes
  • @cedrikewers made their first contribution in
  • #2410 f1a3475 Thanks @sailist! - Fix sporadic "model is not configured" errors when starting kimi web, caused by the background provider-model refresh transiently clearing the model catalog while…
  • #2400 1f3f5da Thanks @7Sageer! - Preserve the assistant's partial output when a turn is interrupted with Esc, and remind the model that the previous turn was deliberately interrupted.
  • #2415 5c0ec29 Thanks @wbxl2000! - web: Enable Monaco-based highlighting for code blocks, and fix line numbers overlapping or drifting out of alignment in fallback-rendered code blocks.
  • 0fdc38704edcab171f0e1b81a5441ceed5055a40 MUL-5149: add agent-generated Chat quick actions
  • c3f5df8bf4381f9855d9f2ae672cdb3cc4841b0c MUL-5492: fix timeline cap dropping newest entries + stop double-broadcasting descriptions
  • cac453fe01193bf1c41b43a6f86c1e77e20d8b21 MUL-5505: fix(selfhost): one source of truth for the backend host port
Open Code Review
  • feat(viewer): add review comments display to session detail page
  • fix: honor per-file review terminal states
  • fix(vscode): force-kill unresponsive reviews
  • Added slash-command and skill expansion to print mode, so a headless run such as -p "/my-skill review this diff" now resolves and applies the skill instead of sending it as literal text, with…
  • Improved interactive startup so a slow or hanging MCP server no longer stalls the first agent turn, loading MCP servers in the background for the interactive session while headless and one-shot…
  • Improved permission grants so a pattern approved at a prompt is recorded for the rest of the conversation, letting later commands that match it run without prompting again.
  • feat: add Yunxiao sync preflight
  • fix: align Yunxiao status names and move project config to repo variables
  • fix(plugin): preserve V7 session defaults

Jul 30, 2026

  • Release Date: July 30, 2026
  • > Patch release. This tag rolls up the ~1,000+ PRs merged since v0.19.0 into a stable tagged release for downstream consumers (Docker images, hosted deployments, fresh installs).
  • Since v2026.7.20 (v0.19.0, July 20): ~2,789 commits · ~4,748 files changed · ~442,000 insertions · ~392,300 deletions on main.
  • fix: preserve authoritative security scan targets
  • fix: document CLI help and deep-scan configuration
  • fix(windows): preserve sandbox override compatibility
  • agent-framework-core, agent-framework-foundry-hosting: Add bounded, in-memory archive skill discovery for MCP sources and expose archive controls through FoundryToolbox
  • agent-framework-core, agent-framework-foundry-hosting, agent-framework-hosting, agent-framework-hosting-responses, agent-framework-hosting-telegram: Add reusable session stores and persist…
  • agent-framework-a2a, agent-framework-ag-ui, agent-framework-anthropic, agent-framework-azure-ai-search, agent-framework-azure-contentunderstanding, agent-framework-azure-cosmos…
  • 支持从 cc-switch 一键导入 Claude Code Provider,并从 OpenAI 兼容接口获取模型列表。
  • 模型目录改为缓存优先、后台刷新,Provider 不可达时不再阻塞桌面端启动。
  • Slash 命令按用途分组;运行位置选择、发送按钮、会话标签和设置侧栏交互进一步简化。
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
  • fix(mcp): pin mcp dependency below v2.0.0 (8e6b33f4b)
  • fix(mcp): deduplicate inverted-index postings (0e517d627)
  • perf(mcp): hydrate search snippets concurrently (fe4cbb948)
Tutti
  • The default desktop agent has been switched to Tutti Agent; new users will use this by default.
  • The enabled state of the Tutti Agent is now persisted across sessions.
  • Fixed a crash loop on startup, making the app more stable.
  • live: Support asynchronous non-blocking tools (c299420)
OpenHands
  • feat: enable/disable an installed MCP server from its card
  • fix: exclude disabled skills from agent context
  • fix: regenerate favicon PNG/ICO from favicon.svg for consistent design
  • The Windows source installer now handles dependency paths longer than legacy MAX_PATH while moving and cleaning its staging tree.
  • The extracted source is renamed to a shorter build path before npm ci creates deeply nested dependencies.
  • Temporary-file cleanup can no longer hide the original installation error.
  • Discover available Modal models automatically. (@devennavani)
  • Prevent adding the same attachment more than once.
  • Always show the new session button.
  • npm: bash npm install -g base44@latest
  • Homebrew: bash brew install base44/tap/base44
  • Version: 0.1.7 NPM Tag: latest
  • Smallest AI: Added SmallestTools, a text-to-speech toolkit for Smallest AI with text_to_speech (returns audio as a ToolResult artifact, optionally saved to disk) and get_voices.
  • AgentOS: Added GET /metrics/refresh/status to observe background metrics refreshes
  • Reports idle, running, completed or failed with started_at, finished_at and error, so clients can poll for completion instead of timing out silently.
  • The Explorer gets project-wide filename search: a persistent search panel at the top of the Explorer column, plus an @-mention dropdown in the chat input sharing the same stream.
  • A search hit can be added straight to the chat as a project file reference with one click.
  • The skill detail page now includes a full file tree and content viewer, with SKILL.md pinned and auto-selected.
  • Channel handlers now receive a 4th argument: ChannelHandlerContext (use ctx.mastra to access the resolved Mastra instance).
  • Channel handlers now receive a 4th argument: a ChannelHandlerContext carrying the resolved mastra instance. Custom handlers can read ctx.mastra.
  • Added exact metadata filtering to message history queries across Memory APIs and supported storage providers.
  • Fixed the Linux Debian package
  • Rebuild Go binaries with Go 1.26.5 and add kubectl build
  • ROB-678: evals for Slack channel-history context recovery
  • ROB-635 ROB-674 Reduce system prompt and tool description sizes (~50-60% fewer prompt tokens)
  • For the installer path, run CodeWhaleSetup.exe; it installs codewhale.exe, codew.exe, and codewhale-tui.exe under %LOCALAPPDATA%\Programs\CodeWhale\bin and adds that directory to the current-user…
  • Extract the archive for your machine: codewhale-windows-x64.zip or
  • Run install.bat (copies to %USERPROFILE%\bin)
  • b13657be71088f35d71425c78cff0d9db0733fed MUL-5493: feat(chat): add a visible follow-up queue
  • da7451843be302a01e66fa9c88ddf059f5d21480 MUL-5494: feat(transcript): render tool events as diffs, content and terminal output
  • 9072cef12cfb0fd74f3ef56b80c6e9ec99930729 Revert "MUL-5493: feat(chat): add a visible follow-up queue"
  • See where your context window goes. The context-usage indicator now breaks the window down by category — system guidance, conversation, tool activity, connectors, and skills — using a model-aware…
  • Never lose track of a finished task. A terminal result from an agent turn now marks its conversation unread unless the focused window already shows it, and a blocking approval request asks for…
  • Find and browse files faster. The project file library gains source-scoped filename search, compact list and grid views, and a large expand modal that preserves scroll position, active filter…
jcode
  • The decorative idle animation is now turned off for all users, including existing configs, via a one-time migration; re-enable it anytime with display.idle_animation = true
  • Ctrl+R reverse history search now behaves readline-style
  • H1/H2 markdown headings render visually larger via underline
  • Published with npm provenance: Republished through the n8n-nodes-mem0-cd.yml GitHub Actions workflow so the package carries a signed provenance attestation.
  • Skill-creator built-in skill: a guided flow to author, update, and delete your own skills
  • Browser sign-in can target a custom login domain for self-hosted or private-gateway deployments
  • Session stats now track provider cache-hit (cached) tokens
  • feat(provider): add Vultr API actions
  • docs: clarify hosted OAuth and expand localized READMEs
  • feat: add Excalidraw MCP provider
  • feat(agent): add token-cost budget guardrails to the review path
  • feat(action): add fail-open category/severity publication controls
  • feat(allowlist): add Prisma schema review support
  • #2365 fa2c5ce Thanks @7Sageer! - Add support for plugin-contributed custom agents, discovered automatically and available for sub-agent delegation.
  • #2314 02d77b2 Thanks @7Sageer! - Allow enabled plugins to contribute agent system-prompt instructions through systemPrompt or systemPromptPath in kimi.plugin.json, effective on both agent engines…
  • #2232 efac96c Thanks @7Sageer! - Support Markdown-defined custom agents on agent-core.
  • 0.6.16 changelog
  • Show provider logos in model menus
  • [feat] Add configurable silence removal controls
  • Surface tool failures instead of reporting them as success
  • Emit FlowFailedEvent when a flow execution fails
  • Implement progressive disclosure for skills
  • feat(usage): add per_request_input_tokens_limit to UsageLimits
  • Refresh KnownModelName from Gateway probes and stable Google image IDs

Jul 29, 2026

  • Set an agent's thinking level from the CLI
  • Switch projects while creating a workspace with ⌘P / Ctrl+P
  • Open a project or workspace folder from the sidebar
  • lossless: factor shared directory prefix in the grep search fold (7dc9a97)
  • metrics: record per-extension token savings (02eb90f)
  • opencode: ship the transport plugin in pip installs (f54f04f)
  • feat: always show the LLM selector; move agent-profile switching to the tools menu
  • feat: add persistent agent memory toggle
  • feat: allow overwriting a secret value from the edit form
  • Credential export for external clients — pi auth print-api-key and pi auth print-bearer-token export configured credentials with automatic OAuth refresh and minimum-validity enforcement.
  • Headless OpenRouter sign-in — Complete /login over SSH by pasting the redirect URL or authorization code when the loopback callback is unavailable. See OpenRouter.
  • Claude Opus 5 on GitHub Copilot — Use Claude Opus 5 through GitHub Copilot with adaptive thinking and a 1M context window. See GitHub Copilot.
Buzz
  • feat(cli): mirror Desktop mention delivery (7adc46268)
  • fix(desktop): deduplicate relay outage notification (66e705492)
  • fix(desktop): reconcile thread arrivals at bottom (b42a8d447)
  • Stable agent event message identity
  • Tab completion and switch provider for model command
  • Support latest Gemini models
Skill Recorder
  • The Windows source installer now saves a Skill Recorder (Source) shortcut
  • The installer finishes with a confirmation that lists exactly where both
  • Shortcut creation is no longer fatal. If Windows refuses to write a shortcut,
  • [embedding] and [rerank] are now soft runtime dependencies — EverOS boots and serves requests with only [llm] configured.
  • everos cascade backfill CLI command — three-phase interactive backfill (vectors → clusters → skills, or --phase all) that upgrades Tier-1 rows to Tier-2 once [embedding] is configured.
  • LanceDB schema v2 — the six business tables (episode, atomic_fact, foresight, agent_case, agent_skill, knowledge_topic) now allow vector NULL, so cascade can write rows without vectors when…
  • docs: changelog and blog post for v0.8.5
  • Fix within-batch cosine similarity for non-unit embeddings
  • chore(deps): bump pillow, gitpython, pyasn1 (clears 62 high-severity alerts)
  • Improved branch picker filtering, grouping, branch creation suggestions, and remote-provider icons. (#61215; thanks chirivelli)
  • Improved LSP Logs by showing request durations alongside RPC responses, making slow language server behavior easier to diagnose. (#61018; thanks aviatesk)
  • Improved solo diffs to show the full file by default while retaining Git change indicators in the scrollbar. (#60989; thanks chirivelli)
  • feat: implement langsmith connector for code brains
  • fix: tolerate malformed slack and x array config
  • fix: preserve contested local-wiki claims
  • Markdown files work as attachments in generate — pass .md files (docs, specs, notes) directly when generating tests, instead of converting them to another format first.
Omniget
  • Seven features that were written, tested and unreachable now have a path from the app to them.
  • Twelve modules had shipped as tested logic with no Tauri command and no call site outside core/ — code that existed but nothing could invoke. Seven are now wired.
  • Failed downloads explain themselves. A failure in the queue is translated into a cause and one action: import cookies, retry, update yt-dlp, free disk space.
  • Qwen Cloud: new provider for Individual Token Plans with 5-hour and weekly rolling windows. Thanks @umutkeltek, and @Yach0 for the API investigation!
  • ZoomMate: new provider with credits, session history, and pacing, using host-scoped cookie routing. Thanks @weddle!
  • Alibaba: Personal/Solo Token Plan variants for mainland (Bailian) and international (Model Studio) accounts. Thanks @LeoLin990405 and @halilertekin for the investigations!
  • The per-conversation workspace file tree is replaced by a project-level Explorer.
  • Tree entries now sort directories-first, like every other file manager.
  • A progress ring next to the send button shows how much of the agent's context window is used (e.g.
  • Immutable artifact versions and provenance. Each generated output can now be inspected as a specific version with checksummed content and linked code, execution, input, environment, branch, and…
  • Message branches for alternative research paths. Edit and resend a completed prompt to explore a different direction without deleting the original downstream conversation.
  • Model and reasoning effort in one picker. The composer now keeps model choice and the selected model's supported reasoning levels together, with live synchronization across supported agent frameworks.
  • cf4114cd5d4cdfca421a62d96b158ebc45caa4fc MUL-5396: validate agent concurrency limits
  • 77907db1e3b30759d02521d3ce244f5a92ca1dd4 MUL-5406: propagate non-ErrNoRows scope DB errors
  • 30318b79bce661e3d00f2da2ab943ab5b435a9cf MUL-5426: fix(daemon): retire sessions whose history the provider refuses to replay
  • Channel reaction tools (add_reaction, remove_reaction) are no longer automatically injected into a channel-bearing agent’s toolset; add them explicitly if needed.
  • Added a readOnly option to NativeSandboxConfig to restrict the local sandbox's working directory to read-only access on macOS (Seatbelt) and Linux (Bubblewrap).
  • Added resolveThreadId to ChannelConfig: resolve the internal Mastra thread id before a channel thread is created, mirroring resolveResourceId.
  • 版本更新:发布 OpenClaw 本地插件 v2.0.12。
  • feat(sandbox): support native host paths in path grants
  • fix: redact Realtime audio format diagnostics
  • fix: redact Blaxel unmount paths in logging
BrowserSkill
  • Fix browser keepalive reconnect
  • fix(extension): fix early control return and preserve request-help handoff across tabs
  • feat(vom): add VOM as BrowserSkill observation layer and integrated into extension pipeline.
  • Mistral provider - Full Mistral SDK support (streaming, multi-modality, structured outputs{
  • chore(deps): bump the uv group with 20 updates
  • chore(deps): bump the github-actions group with 6 updates
  • #2255 67dd031 Thanks @he-yufeng! - Add a customizable footer status line, configured via [status_line] in tui.toml.
  • #2313 de0ba9d Thanks @starquakee! - Stop the turn after repeated invalid tool calls instead of retrying indefinitely.
  • #2147 29783e4 Thanks @wbxl2000! - Show a quota note after installing official plugins that bill against plan quota (such as Kimi Datasource).
  • Add Claude Opus 5 (claude-opus-5) support
  • Add OpenAI Responses API reasoning.context support (default all_turns) for the gpt-5.4/gpt-5.5/gpt-5.6 families
  • Preserve arbitrary RequestUsage fields across serialization
  • Workspace browsing: browse the files under the workspace directory from the panel, with search and download to local.
  • File preview: HTML pages, documents, and images open directly in the side panel, and file paths in a reply are rendered as clickable cards.
  • File references: drag a workspace file into the conversation, or use @ in the input box to reference files and directories and pull them into the context.
WrenAI
  • finish wren-engine → wrenai rename cleanup (0733d37)
  • pydantic: skip non-dict models in list_models (507bda6)
  • sdk: raise the wrenai floor to 0.13.1 (3615a8d)

Jul 28, 2026

  • This set of releases tightens up the shipping pipeline and keeps the release branch moving smoothly—so builds stay green and delivery stays fast. 🎉
  • One promotion also required a careful conflict resolution pass to preserve the already-cut release versions while still bringing forward the full set of changes from main. — Thank you @M3gA-Mind!
  • Voice mode keeps up with the work: Hear updates when something actually changes instead of getting the same message every few seconds.
  • Image generation gets a proper home: Choose image generation as a provider, just like you choose a model. It is no longer hidden inside a separate mode.
  • More models to choose from: Claude Opus 5, GPT-5.6 Sol, Kimi K3 through Fireworks, and Poolside are now available for different kinds of work.
  • feat(web,ui,desktop): add /design component showcase and standardize tooltip style
  • feat(desktop): filter and sort projects in the dashboard sidebar
  • feat(desktop): add PostHog usage tracking for terminal rich input
  • fix: increment previous_content with content streamed
  • fix: add default fallback to instantiation.py and removed default ollama base url from components
  • fix(tracing): detach orphan spans before database flush
  • bb desktop 0.34.0
  • release(langgraph): 1.2.10
  • chore(deps): bump jupyterlab from 4.5.9 to 4.5.10 in /libs/langgraph
  • chore(deps): bump setuptools from 80.9.0 to 83.0.0 in /libs/langgraph
OpenCode
  • Restored compatibility with legacy MCP SDK clients.
  • Fixed a Solid cleanup crash that could break navigation in the desktop app.
  • Fixed home session loading so the session list can update without suspending the whole page.
  • fix(release): recover npm publish after tombstoned 1.6.0
Mistral Vibe
  • Crash when @-mentioning a file, which aborted the turn
  • #1987 ad015c2 Thanks @mattzcarey! - Update the MCP dependencies to stable @modelcontextprotocol/client@2.0.0 and @modelcontextprotocol/server@2.0.0, and update the retained SDK v1 compatibility…
  • #1982 e983026 Thanks @mattzcarey! - Fix AI SDK v7 telemetry, which produced spans with no token counts, no finish reason, no tool results and zero durations.
  • Composer and slash palette are gone — the new options-first surface replaces them with a shared panel family.
  • Question panel shows your picks and lets you act on them in bulk — visible selections, real multi-select, and a » Go with recommendations batch action (reachable via Tab) make answering faster on…
  • Design check-in is phase-aware — the headline updates per phase, labels switch between "Design view" and "✎ Edit", and a "Continue to next" prompt appears at the right moment instead of leaving…
  • Add WaitTool for pausing on long-running jobs.
  • Fix FileWriterTool writes and address rough edges in file tool.
  • Mark E2B_API_KEY as a required env var for E2B tools.
  • feat(invites): add use-limited invite links (d500c2d5c)
  • fix(node): bump Buzz-supplied Node runtimes past OpenClaw's >=24.15.0 floor (98a7b1334)
  • fix(desktop): preserve thread anchor through layout reflow (9810d8545)
  • fix(pptx): notes slides get the package relationships they require
  • fix(cli): decode piped stdin as UTF-8 without mutating the console
  • Preserve Excel drawing hyperlinks and grouped shapes during round-trip
  • Low-risk, one-off permission options (allow once, reject, modify) now submit with a single click — no more pick-then-confirm two-step.
  • Left-clicking the tray icon now toggles the main window show/hide, Discord/Slack style. Right-click keeps the context menu; macOS keeps its platform convention.
  • The Enabled tab in Assistant Settings now shows a quick-chat button on each row, so you can jump straight into a conversation with any assistant from the settings page.
WrenAI
  • ask: guided template's recall step uses a valid CLI option (f242be4)
  • athena: strip trailing semicolon on unlimited query path (06e4c42)
  • bigquery: push LIMIT into SQL and strip trailing semicolon (0a25df3)
  • aa110aac9c1f5bf30797d962c0a3a0977650c5f2 MUL-5088: import repositories from GitHub App
  • c271f80999ef1017e0be300a94a6f85255b72d55 MUL-5370 fix: label stalled skill-bundle downloads, align failure-reason copy with the backend taxonomy
  • 18adb20c149fbdf30540a39fd8eb945187cec94b MUL-5391: unify upload placeholder UI
  • npm: bash npm install -g base44@latest
  • Homebrew: bash brew install base44/tap/base44
  • Version: 0.1.6 NPM Tag: latest
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
  • The daemon now ships a default idle timeout of 1 hour: after an hour with no commands or dashboard input it saves configured restore state, closes the browser, and exits, so integrations that die…
  • Fixed tab recovery and selection to avoid daemon hangs on Memory Saver-discarded tabs by selecting a live renderer on CDP connect, reviving tabs on switch or after close, treating dialog-blocked…
  • Fixed a11y selector errors to report invalid CSS selectors cleanly instead of exposing raw browser evaluation stack traces in text and JSON output
  • Typing in the input line stays smooth while a turn is streaming: buffered keystrokes are coalesced into one frame and user input now beats server events to the loop
  • Remote-mode clients drain bursts of buffered terminal input (fast typing, key repeat, scroll wheels) into a single render, matching the local loop
  • Both interactive loops poll terminal input ahead of timers and server/bus chatter, so heavy streaming can no longer starve keystrokes
  • Workflow tool nodes now support multi-select dropdown inputs, allowing users to choose multiple values from a predefined list when configuring tool parameters.
  • You can now click a node_id in workflow run logs or error messages to locate and highlight the corresponding node in the workflow editor canvas.
  • The block selector panel now features improved keyboard navigation, category previews, and a more intuitive search experience when adding nodes to workflows.
  • Add headers and retry_after to ModelHTTPError, populate from all provider SDKs
  • Re-assert absorbed external cancellation at run step boundaries
  • Update FastMCP dependency range to <4
  • feat(allowlist): add Bicep (.bicep) support
  • feat(allowlist): add HCL/Terraform (.hcl, .tfvars) support
  • feat(mcp): support remote MCP servers via Streamable HTTP
  • Print mode (-p / --print) now supports structured, machine-readable output via the --output-format flag (text (default), json, or stream-json), so headless runs in CI, eval harnesses, and scripts…
  • Added the stream-json output format: a strongly-typed NDJSON event stream that emits typed init, step_update, and terminal result events with a stable, closed-vocabulary step_type discriminator…
  • Added the --json-schema flag to enforce a custom JSON schema on the structured output, accepting either an inline schema string or a path to a schema file.

Jul 27, 2026

  • ⏰ Automations (Scheduled Tasks) — a new /tasks page to create, edit, and run recurring agent tasks, with model and reasoning-effort selectors, a "Run now" action, live relative next-run times, and…
  • 📁 First-class Projects — organize work into real project entities: create, rename, and delete them in the sidebar, file sessions in, and give each project default session settings…
  • 🎙️ Voice dictation in the composer — optional server-side transcription (omnigent[dictation]) works in Electron, Firefox, and Chromium with live streaming partials and audio that never leaves your…
  • chore(deps-dev): update redis requirement from <8 to <9
  • chore(deps-dev): update azure-identity requirement from >=1.21.0 to >=1.25.3 in /mcp_server
  • chore(deps-dev): update google-genai requirement from >=1.72.0 to >=2.8.0 in /mcp_server
Strands Agents
  • feat: add context manager class design doc (cb3bffba)
  • feat(ts): add agent-as-tool delegation (a2761fd2)
  • fix(context): legacy file storage accepts bare filenames and stems (2828e442)
Strix
  • fix(tls): replace raw urllib with requests for external HTTPS calls (frozen-build cert failures)
  • chore: release v1.4.1
  • Platforms. Seven targets — macOS (Apple Silicon, Intel), Linux
  • Guided setup. ironclaw onboard provisions the config, the encrypted
  • Model providers. 26 providers in the built-in catalog, including NEAR AI,
  • Added Claude Opus 5 support for Anthropic and Amazon Bedrock BYOK providers.
  • Fixed an issue where hover documentation could have too many line breaks.
  • Click elements or select text to attach numbered annotations
  • Chat with the agent from a side rail while it works in the terminal
  • Approve plan / Request changes buttons map directly onto /plan's CONFIRM gate
  • #1557 447013d Thanks @mattzcarey! - Add MCP SDK v2 client and server support. MCPClientConnection now uses the exact-pinned @modelcontextprotocol/client@2.0.0-beta.5.
  • McpAgent is deprecated and feature-frozen as a stateful SDK v1 path. New servers should use an SDK v2 factory with createMcpHandler from agents/mcp/server.
  • Passing an SDK v1 server to the overloaded createMcpHandler is deprecated for removal in the next major release. Move the server to an SDK v2 factory.
  • Manage workspace scripts from the CLI and agent MCP tools
  • Copy terminal IDs from terminal tab menus
  • Long Markdown lines wrap by default in the file editor
  • fix: dont send default params in gemini genconfig
  • Fix tool name in flash prompts: replace_file_str -> replace_file
  • fix: handle file:// urls in _extract_start_url
  • fix: eliminate parallel Rust test flakes
  • feat(cli): add --output json/yaml to sandbox get, status, and sandbox create
  • refactor(tui): default create form state
  • @mastra/braintrust: Bundled Braintrust SDK upgraded v2 → v3; BraintrustExporterConfig types narrowed (logger/span shims) and root_span_id semantics differ in v3.
  • mastra build now deploys in one step for push-style deployers. Deployers can opt in with the new deployOnBuild flag on the deployer contract, and the build runs their deploy() right after bundling.
  • The 'workers' option on the Mastra class now merges with the default workers instead of replacing them. Passing custom workers (e.g.
  • Response tool routing fix — Native Responses output text is now routed through the normalized response-tool executor instead of ending the monologue directly, preserving active goal overrides…
  • Streaming resilience — Partial tool arguments no longer raise errors during streaming.
  • Prompt JSON fence handling — Fenced JSON examples in prompt Markdown are now stripped of fence markers at render time so the assembled system prompt matches the required wire format.
  • PDFs can now drive test runs — a PDF supplied as run_config.source is ingested, converted to a structured projection, and delivered with a complete content inventory so the AI knows exactly what…
  • Embedded images inside PDFs are fingerprint-verified — each image extracted from a PDF is hashed at ingest, its fingerprint is checked through staging and delivery, and its origin page is tracked.
  • Payload size is enforced at the gate — if a PDF's serialized payload exceeds the size contract, it is blocked before reaching the runner rather than failing mid-run.
  • Native context compaction. Claude Code and OpenCode sessions now invoke the framework's own compaction mechanism — automatically at 90% context usage, or on demand from the context-indicator…
  • Dark mode. A warm, low-chroma dark theme is available under Settings → General, applied synchronously before first render (no flash), covering the shell, transcript, and all renderer palettes.
  • Whole-window find. Cmd/Ctrl+F opens an overlay search bar over the entire workspace — agent transcripts, notebook output, and rendered pages — with a live match counter, next/previous navigation…
  • /remote is a new front door for reaching a running session from another machine over the WebSocket gateway
  • Destructive shell commands like recursive deletes now pause for an explicit reflection step before running, closing several bypass routes found in review
  • The desktop preview gained selectable transcript text, an Alt spatial session overview, live reasoning and tool-intent streaming, and much smoother scrolling
  • Captain FAQ suggestion review API and interface.
  • WhatsApp health monitoring, BSUID messaging, and clearer outbound call errors.
  • Intercom import retry and reliability improvements.
  • e52b50658a66d09bffed126e34116ad826c03623 docs(changelog): add v0.4.12 release entry (2026-07-27) (MUL-5364)
  • 4931ee6f7aa586c998dff0bcde498a1a02bf6ed4 docs(chat): correct the chat sandbox description (MUL-5287)
  • 3d4c5c7da2eb11b2238061a3bdc15639e8488275 feat(cli): add 'multica agent copy' to fork an agent across runtimes (MUL-5279)
  • AgentOS Tools: Added AgentOSTools to report on AgentOS usage, latency, failures, schedules, evals, components, and pending approvals.
  • Traces: Added latency and error stats grouped by agent, team, workflow, or endpoint, plus tool and model call stats. Implemented for PostgresDb and SqliteDb.
  • Default changed to kimi-k3.
OpenCode
  • Remove the extra titlebar inset in fullscreen on macOS.
  • Keep command palette entries from reappearing incorrectly when shadowed commands are removed.
  • Add scrolling to the project selector dropdown when the list is long. (@david1gp)
  • Add JSON logging format support for log scrapers
  • ROB-602 Langfuse support in OpenTelemetry tracing
  • fix holmes operator health checks when using vertex_ai/gemini
  • 批量反射评分功能:新增批量反射评分功能以优化性能。
  • 修复 V7 会话默认设置:确保 V7 会话的默认设置在插件中得以保留。
  • #2192 7799bd7 Thanks @sailist! - Hold per-agent runtime state of the experimental engine in the agent-scope state container, so it is observable in one place and disposed with the agent.
  • #2119 f06eb5c Thanks @pvzheroes125! - Allow hosts to defer registered user-tool schemas until needed. Set disclosure: "deferred" when registering a tool.
  • #2192 7799bd7 Thanks @sailist! - Instantiate every registered service eagerly at scope creation on the experimental engine, following the dependency graph automatically, and drop the…
  • Rewind, fork, replay, and jumpToFrame run an effect-boundary
  • The diagnosis follows you everywhere. rewind and timetravel
  • The journal is observable. The SmithersEvent union gains
  • Added agents.tool.ProgrammaticToolCallingTool, which lets supported OpenAI Responses models generate JavaScript to coordinate eligible tools.
  • Added the public agents.decorators module and the shorter @tool alias alongside the existing function and guardrail decorators. Function tools now also support async callable objects.
  • SDK configuration now consistently accepts either typed settings objects or dictionaries across agents, runs, models, sessions, sandboxes, and voice pipelines, with validation for unknown settings.
  • feat(provider): add Mux video
  • feat(provider): add AgentMail (QQ) agent_qq provider
  • fix(auth): require admin token for action-run elevation
  • Intercept extra TCP ports beyond 80/443 (experimental). New env var OPENSANDBOX_EGRESS_MITMPROXY_EXTRA_PORTS (comma-separated) appends extra destination ports to the transparent-mitm REDIRECT…
  • npm scoped packages no longer 403 under Credential Vault. A single-layer %2f (e.g.
  • DNS survives deny.always overlays containing 127.0.0.0/8. In dns+nft mode, DNS redirected to 127.0.0.1:15353 is now accepted before the deny sets, so a loopback deny rule no longer breaks sandbox…
  • NLIP — Integration for the Natural Language Interaction Protocol.
  • ACP — Support for the Agent Client Protocol, for driving CLI coding agents, together with an MCP tool gateway that exposes AG2 tools to those agents.
  • MCP-UI — Support for the MCP UI extension.
  • 「纸·墨·印」全量 UI 重设计:6 套主题(纯白、纸墨、经典暖色、青瓷、墨夜、墨夜蓝),采用源色块加单一语义映射层的两层 token 结构;标题改用自托管的 Noto Serif SC 衬线拉丁分片,中文回退系统衬线;圆角、阴影、动效阶梯统一定稿。外壳、零态、活跃会话、composer 全部弹层、技能市场、定时任务、设置全分区、Trace、终端和 ⌘K…
  • 桌面端组件库:新增 components/ui 原语层(24 个组件)与 components/composite 共享部件层,替换此前散落的原生控件——362 个原生 button 只有 22% 复用率,图标按钮存在 15 种圆角、12 种尺寸,41% 没有焦点环,10 个没有可访问名称。组件复用率现为 75%,剩余部分是…
  • 品牌标记矢量重建:应用图标从 120KB 位图重建为矢量几何,可以随主题重新着色——两个 C 取墨色,印章条、光标和星芒取朱色。所有消费点按尺寸减件:星芒在 40px 以上、光标在 24px 以上,C 与印章条始终保留。应用图标本身保持一份固定艺术稿(Dock 与任务栏由系统绘制,不感知应用内主题),应用内的标记则是跟随主题重绘的矢量。

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.