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: May 25 – May 31, 2026

123 stable releases from 67 tracked agents and frameworks in the week of May 25, 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.

May 31, 2026

  • Custom command keybindings now accept an optional description field to provide user-defined descriptions shown in the keybind help panel instead of the default 'custom command' label.
  • The OpenCode integration no longer treats session.created or session.updated plugin events as idle signals, so active sessions stay marked working until OpenCode reports session.status or…
  • New interactive panes now use login-shell startup on macOS by default so Homebrew and other login PATH setup is available, with terminal.shell_mode = "non_login" as an opt-out.
  • 右侧 Workbench + 内置浏览器:右侧面板支持 Files / Browser 切换,带地址栏、加载状态、child webview 生命周期和窗口边界同步。
  • 预览与 Open With 更完整:助手输出、变更文件、workspace 文件、localhost、本地绝对路径和 /preview-fs / /local-file 都能走统一预览或 Open With 动作。
  • 网页圈选编辑链路:Browser 页面支持元素选择、截图、编辑气泡、标注区域和干净的 composer 回填,适合把网页局部修改请求直接交给 Agent。
  • Extract codewhale-windows-x64.zip
  • Run install.bat (copies to %USERPROFILE%\bin)
  • Add %USERPROFILE%\bin to your PATH
  • #1334 a1b4427 Thanks @omeraplak! - fix: prevent unhandled rejections when aborting Agent.streamText() streams
  • Aborting a consumed streamText() stream no longer emits unhandled rejections for unconsumed result fields
  • Callers using only toUIMessageStream(), toUIMessageStreamResponse(), fullStream, or textStream do not need to attach defensive .catch() handlers to text, usage, or finishReason
  • c569ad313c5304fd832339533cde15f0b31a668b Merge branch 'main' into fix/ua-client-hints-consistency
  • e6bbabc31d668480b58a924ec2649ca93e458f55 Merge pull request #575 from pinchtab/feat/add-beyond-viewport-screenshot
  • c20ebac6ac1d26cbe21fb24033f3233f29d4cd49 Merge pull request #576 from pinchtab/feat/add-capture-endpoint
  • fix(ci): add v1 branch to workflow triggers
  • refactor: update tool components for improved styling and functionality
  • SKY-10578: Add credential folders (backend) (7e6ea829d)
  • fix(SKY-10622): clean up child entities on workflow run timeout (66d1fe98d)
  • fix(frontend): replace react-github-btn with static link (SKY-9724)
  • Failed-message error tips now keep their structured, classified UI even on live streaming responses and build-task failures, instead of occasionally falling back to a plain error message.
  • Backend startup failures now capture far richer diagnostics — process ID, health-check timing, fetch errors, server-listening detection, and a TCP probe — making it much easier to understand why a…
  • Packaged installs where the runtime exists but the backend binary is missing are now correctly identified as incomplete installations, with install paths, runtime manifest details, and recent…
  • feat(live_component): coco.auto_refresh + unify error-handling channel
  • fix(serde): register _frombuffer under numpy 1.x and 2.x paths
  • fix(live_component): scope LiveComponentOperator to one process_live invocation

May 30, 2026

  • Dolt 2.0.7 or newer is required.
  • Beads bd v1.0.4 is the supported version for this release. Gas Town now fails fast if it detects a newer unsupported bd version, rather than risking schema drift against shared workspaces.
  • Release archives are built CGO-disabled across the GoReleaser matrix.
  • Better memory recall: improved search relevance and ranking
  • Faster ingestion and syncing of memories across devices
  • New memory search UI improvements for quicker results
  • Inline document annotations and comments - Issue documents now support inline, revision-aware annotation threads with comments and stable anchor snapshots, so agents and operators can discuss a…
  • Company skills CLI and catalog management - Skills are now first-class: install, reset, audit, export, and assign company skills with a new CLI and a board UI, backed by a packaged skills catalog…
  • Hide projects and agents from your sidebar - User-scoped resource membership lets each user leave projects and agents they don't want cluttering their sidebar while keeping every resource…

May 29, 2026

  • Named startup sessions - --name / -n sets the session display name before startup across interactive, print, JSON, and RPC modes. See Naming Sessions and Session Options.
  • Clickable file tool paths - built-in file tool titles render OSC 8 file:// hyperlinks when the terminal supports them, including supported tmux clients.
  • Exported convertToPng for extension authors
  • Claude Opus 4.8 now available: Anthropic's latest Opus model is live in the hosted catalog and selectable across all surfaces
  • Gemini 3.5 Flash on the hosted service: the model was already selectable in the local UI but a missing platform-side rate card returned a 400 for every cloud request.
  • Sleep policy 1-hour and 3-hour presets: two shorter idle-sleep durations join the sleep policy control in Settings alongside the existing 1-day, 3-day, and 7-day options so assistants can be tuned…
  • (openai realtime): add status_details to incomplete response logs
  • feat(cartesia): add ink-2 stt
  • fix(ipc): bound AgentSession.aclose() during job shutdown
  • e14d47a8135cef6950d6a337a8f7e3c62157bc27: fix(noninteractive): crush run reliability in client/server mode (@meowgorithm)
  • b7f4ad6cf61bcca11e052f19c2f2b653aaf818be: fix(opencode): fix qwen3.7-max on opencode (@andreynering)
  • a4181d6d4ecc3a246ae949784b115f6c970bd105: fix(scrollbar): fix track position calculation (@andreynering)
  • Revamped agent management: new endpoints, connection naming, unregister/deactivate flows, and SDK access to agent configurations.
  • Major visualization rewrite: pipeline-aware Story layout, schema view, clearer legend and UI controls, and better rendering of edges/labels.
  • New/updated graph dataset handlers (local Neo4j support) and improved edge ingestion and fallback labeling.
  • Added pane copy mode at prefix+[ with keyboard navigation, visual selection, and clipboard yank support.
  • Added foreground_cwd to pane and agent API/CLI responses so integrations can inspect the active foreground process directory without changing the existing pane/workspace cwd semantics.
  • Added read-only agent_session metadata to pane and agent API/CLI responses when official integrations report native session references.
  • User Resources — a persistent, per-user file library with MD5-deduplicated storage and a virtual path filesystem.
  • Flow files — per-flow workspace files that sync into worker containers at /work/uploads and /work/resources. Files can be pulled back out of a running container and promoted into the user library.
  • File Manager UI — a reusable tree component with multi-select, keyboard navigation, drag-and-drop, sortable columns, bulk actions, and an overwrite workflow.
  • Opt out of auto-generated checks in action mode (kane-cli#43) — when authoring tests via kane-cli in action mode, the CLI no longer appends its own final-verification check.
  • Confidence-scored element matching — every element-match call now reports a confidence score and the visual cues that influenced it.
  • Richer target descriptions — element targeting now distinguishes load-bearing descriptors (PRIMARY) from supporting visual cues (HINTS), producing fewer ambiguous matches on visually similar elements.
  • docs: refresh landing terminal demo and apply NVIDIA fern theme
  • build(macos): remove unused import of tracing::warn
  • chore: align .python-version with mise.toml
  • MCP servers are now managed per conversation instead of being synced globally to CLI agents, so each chat keeps its own toolset without affecting others.
  • Importing MCP servers from JSON now validates the payload up front — bad URLs, missing commands, or malformed configs surface as clear UI errors instead of silent failures or unhandled crashes.
  • When an agent fails to respond, you now get a readable explanation with a suggested fix, plus retry and feedback actions, instead of a raw error dump.
  • fix: update Docker base images to Trixie and force pull latest images in nightly builds
  • fix: Model handling for tool calling in agents and update IBM models
  • fix(ci): backport bundle rename step to nightly workflow on main
  • enable_system_trust_store config flag to switch the shared SSL context to the OS trust store for corporate TLS / private CA setups
  • MCP HTTP transport now uses Vibe's shared SSL context so it honors SSL_CERT_FILE / SSL_CERT_DIR and the system trust store opt-in
  • API key onboarding and plan-upgrade CTAs now link to the new Mistral Vibe Code extensions page
  • Add Claude Opus 4.8 to the model picker (replaces Opus 4.7)
  • Add market-wide earnings feed: optional ticker, limit parameter, and table formatting for multiple records
  • Update financial_search routing so queries like "latest earnings" or "earnings feed" call get_earnings without a ticker
  • f062cb41d70a5bfb05b06a5bbcc819b1bf129ca5 1
  • 1055e082a427f8e055465cb64456e3271e038fba Add MCP section to config web UI
  • bdaff5cb693f2c5efbc0105c19d9c8459e7dab2c Add pretty_print and disable_escape_html to tool_feedback defaults
Hermes Agent
  • Packaging): ship bundled plugin.yaml manifests in wheel and sdist (827f7f07)
  • @outsourc-e (1 commit)
  • @dparikh79 (1 commit)
  • New documentation-style Help Center layout and layout switcher.
  • Conversation unread counts, sidebar ordering, and badge updates.
  • WhatsApp/Twilio Cloud Calling and voice-call UX fixes.
  • 98c40029155635a0cd7c01d2724d8001ff323d28
  • feat: add Resend email piece @Tosh94
  • feat: add Plausible Analytics piece @Tosh94
  • fix(typefully): replace polling with webhook trigger, fix i18n, add test function @bst1n
  • fix(langchain): recover stale OpenViking clients
  • feat(search) Add lightweight query planner config for intent analysis
  • fix(server): update server mode terminology from 'dev-implicit' to 'dev'
agentmemory
  • GitHub Copilot CLI (PR #534). Full first-class support: plugin/plugin.json manifest, hooks.copilot.json lifecycle hooks for all 11 Copilot events with camelCase payload normalization, agentmemory…
  • Five new MCP adapters (PR #677). Warp, Cline, Continue, Zed, and Droid. createJsonMcpAdapter extended with wrapperKey (Zed uses context_servers instead of mcpServers) and extraEntryFields…
  • /agentmemory/graph/build endpoint (PR #698). Backfills the knowledge graph from existing compressed observations across every session in configurable batches.
  • unsupported media drop feedback toast
  • Show generating preview for in-progress AI media
  • haptic feedback for snap engine
  • Bump dev version: v1.26.0.dev0
  • Add new possible error message to test of LiteLLM providers w/o key
  • Fix typo cleanups across docs, examples, and tests
  • clear 5 high-severity Dependabot alerts on main (cee71e6)
  • Pipecat pipelines are multi-agent compatible by default. The new multi-agent framework (pipecat.workers) turns every PipelineWorker (previously PipelineTask) into a peer on a shared bus that…
  • Added UIWorker (pipecat.workers.ui): an LLM worker that observes and drives a client web UI over the RTVI UI channel — for voice agents that act on what the user is looking at.
  • A voice agent delegates a turn via the built-in respond job; the worker returns an answer for the voice LLM to speak, or speaks it verbatim through the agent's TTS with respond_to_job…
  • 🧪 Agent Evaluations – New evaluation framework for grading agent outputs (autogen.beta.eval).
  • 📁 xAI File API – Implemented file API support in the Beta xAI provider.
  • 🔧 OpenAI Config Cleanup – Omit empty parameters in Beta OpenAI configs.

May 28, 2026

bb
  • bb desktop 0.0.12
  • Claude Opus 4.8 support - Adds Anthropic Claude Opus 4.8 metadata and updates Opus adaptive-thinking coverage.
  • Selective tool disablement - --exclude-tools / -xt disables specific built-in, extension, or custom tools while leaving the rest available. See Tool Options.
  • Headless Codex subscription login - /login can use device-code auth for ChatGPT Plus/Pro Codex subscriptions. See Subscriptions and OpenAI Codex.
  • e764240d9ceb74a626cccdd404afaaa76e128314: feat(api): expose in progress flag on session responses (@meowgorithm)
  • ec85e9ef686c23ce3366c46589cdfabc05cd2199: feat(api): report how many clients are watching each session (@meowgorithm)
  • 6923820a8f60a06a0a25100214871ccfd382e8e1: feat(db): refuse to open a data directory in use by another crush (@meowgorithm)
  • fix: reasoning models reject tool_choice=required; bump to 1.0.2 (closes #503,
  • The Big Refactor — run_agent.py is no longer 16,000 lines — The file at the heart of Hermes — the agent conversation loop — has been reduced from 16,083 lines to 3,821 (-76%), with the extracted…
  • Kanban grew into a real multi-agent platform — 104 PRs end to end — Triage auto-decomposes one task into a tree of sub-tasks.
  • Cold-start perf wave keeps going — another second saved, 47% fewer per-turn function calls — Three new optimization rounds: defer openai._base_client import…
  • STG-1746 Add SEA --version flag
  • Enhance StdioTransport to prevent environment variable leakage
  • Enhance planning configuration and observation handling
  • Declare env_vars on DatabricksQueryTool
  • fix: clarify OpenClaw external install prompt
  • fix: support Grok 4.3 reasoning effort in xAI responses
  • fix(code): keep light theme tokens readable
  • docs(kubernetes): note that Sandbox volumeClaimTemplates is immutable
  • fix(sandbox): use succinct endpoint denial reason
  • feat(docker): add provisioning progress events
  • Re-opening a conversation no longer pays for refetching every long tool output up front — full tool content is loaded on demand when you expand a truncated tool call, so big histories open…
  • Removed redundant warmup and MCP sync work on the message-send path so the first reply in a conversation arrives sooner, especially on aionrs and ACP backends.
  • Switching the model inside a conversation now sticks: leaving and returning to the conversation keeps your chosen model, and the home page dropdown remembers the last choice across ACP, aionrs…
  • AgentSignalContents narrowed to string | (TextPart | FilePart)[] (no longer accepts wrapped BaseMessageListInput shapes); update agent.sendSignal callers accordingly.
  • @mastra/playground-ui Button variants changed: cta/contrast/link removed in favor of default/primary/outline/ghost; ButtonWithTooltip removed (use Button with tooltip prop).
  • PopoverContent no longer forwards onOpenAutoFocus/onCloseAutoFocus (use initialFocus/finalFocus instead).
  • Release Notes
  • Breaking Changes
  • docs(retrieval): add pgroonga to the BM25 backends table
  • docs: add 0.7.0 changelog and release blog post
  • fix(control-plane): add graph_maintenance to operations type filter
  • Full Changelog
  • agent-framework-core: Add HarnessAgent and background-agents harness provider
  • agent-framework-core, agent-framework-a2a: Add A2AAgentSession with referenced task IDs and input-required support
  • agent-framework-foundry: Add experimental prompt-agent conversion and deployment APIs
  • 32e904c017c33eb35f7abb5a9e6e61e2e7aea81c .Net: Bump .NET package version to 1.77.0
  • 3e180c16b3004f8b10720ea5f82c3ad136e59153 .Net: Enable default-on server URL validation for OpenAPI plugins
  • fdc6e68f5892c3f76519d22a770d7e0a12a261f6 .Net: Semantic Kernel -> Agent Framework - Migration/Samples Update (AF 1.0 Compatible)
  • dependencies resolution
  • ignore empty tool messages
  • preserve LiteLLM text with tool calls
  • Auto-deny orphaned tool calls to prevent LLM failures
  • feat: add ModelsLab image generation provider
  • fix: NameError when STRATEGIC_LLM lacks colon separator
  • fix: properly resolve report_type imports in websocket_manager and re…
  • Recall 上下文预算控制 (#71 / #70):新增 recall.maxCharsPerMemory 与 recall.maxTotalRecallChars 配置,按分数顺序裁剪超长条目并丢弃溢出部分,避免长会话因记忆膨胀挤占上下文。
  • L1 / L2 / L3 提示词按用户输入语言自适应:l1-extraction、l1-dedup、scene-extraction、persona-generation 四个 prompt 的自由文本字段现在跟随用户消息语言书写,无需配置 locale。
  • Embedding sendDimensions 可选关闭:新增 embedding.sendDimensions 配置项,设为 false 时省略请求体中的 dimensions 字段,可对接 BGE-M3 等固定维度模型。
  • Release version 3.3.6rc1 with exclusive Rust fastembed backend for …
  • fix mysql datetime type
  • Release version 3.3.6, updating changelog and project files to reflec…

May 27, 2026

  • TUI command on goose-cli
  • TUI diff viewer
  • goose review local code review command
  • (openai realtime): add gpt-realtime-2 model str
  • (healthcare example): remove filesearch and pdf
  • chore(worker): update worker warnings AGT-2909
Mistral Vibe
  • VS Code extension promo link in the CLI banner now points to the renamed mistralai.mistral-vibe-code extension
  • /teleport now uses the new Vibe Code Web sessions
  • Explicit session IDs for automation - --session-id <id> lets scripts create or resume an exact project-local session. See Sessions.
  • RPC bash output can stay out of model context - RPC clients can pass excludeFromContext to bash for commands whose output should not be sent with the next prompt. See RPC mode.
  • More predictable provider retries and timeouts - Codex WebSocket/SSE waits are bounded, and retry.provider.maxRetries controls provider retries instead of hidden SDK defaults. See Retry settings.
  • Add screenshot option to Extract
  • feat(verifier): add evaluator backend facade
  • Workflow: publish eval results
Herdr
  • Fixed macOS server startup with large restored sessions by raising the server file descriptor soft limit, preventing new panes from failing with dup of fd N failed or Too many open files around 40…
  • fix(shell): use app-level hide on macOS close button
  • feat(memory): activate chat memory pipeline — Archivist wiring, LLM recaps, cross-thread STM recall, segment-granular tree, unified compaction
  • feat(memory): #1574 Stage 2 — per-(row,model) embedding storage live (cutover + migration + re-embed backfill + switch UX)
  • feat: add fastuuid package and enhance MongoDB ID handling
  • chore: remove unused cargo arguments in publish-npm workflow
  • Warn users when allowConversationAccess is not configured (>= 2026.5.7)
  • @mastra/slack: removed cards: boolean and formatToolCall from Slack provider/channel config; migrate to toolDisplay: 'text' (for cards: false) or toolDisplay function form (for formatToolCall).
  • Improved agent channels UX
  • Streaming text — opt-in per-adapter streaming flag (boolean | { updateIntervalMs?: number }) that pushes the agent's text deltas into the platform message progressively via the Chat SDK.
AionUi
  • Builtin assistants now show a readonly banner with an inline "duplicate" link, making it obvious how to fork one for editing.
  • The image-generation model dropdown is restricted to known-working models (Gemini image, OpenRouter chat-image, AntigravityTools), with a tooltip listing supported providers — no more silent…
  • The Capabilities tab is renamed to Tools, and speech-to-text settings now collapse when the toggle is off.
  • cab700978b15dbea1324bdebd2e3847fd0706f17
  • [1.71.x backport] fix(native-webpack-plugin): support @vscode/ripgrep >= 1.18.0 in
  • Notes: The minimum required @vscode/ripgrep is now ^1.18.0. Fresh installs already resolved to 1.18.0 via semver, so most consumers are unaffected.
  • refactor: deduplicate shared code across ocsf builders and driver crates
  • fix(python): raise SandboxError instead of FileNotFoundError or KeyError
  • fix(scripts): replace mapfile with bash 3.2-compatible read loop in helm-k3s-local
  • This release patches two security advisories and continues the performance work from v1.14.0 with a persistent token-count cache, plus expanded Dart parsing and Nix support.
  • A crafted --remote-branch value could be passed to git as an option rather than a ref, enabling argument injection (CWE-88, High).
  • Special thanks to @kakashi-kx (Abhijith S) for the responsible disclosure! 🎉
  • docs: add 0.6.2 changelog and release blog post
  • docs: attribute 0.6.2 release post to benfrank241
  • blog: What's New in Hindsight Cloud
CodeBurn
  • CLI update notification. When a newer CLI version is published, a blue banner appears above the star row showing the version and a copy button for the update command…
  • Loading watchdog killed healthy CLI fetches. The recovery loop ran every 8 seconds with no backoff.
  • Multi-day cache key mismatch. selectedDay returned the earliest date instead of nil when multiple days were selected, and startInteractiveSelectionRefresh did not pass the day set to the cache key…
  • Warp Stable release
  • Tauri 2.x Desktop App: New Tauri-based native desktop app for macOS and Windows
  • Coding Mode: New Web IDE with a three-panel layout — file tree, tabbed editor with inline diff review, and a Git panel for branching, staging, and committing
  • Skill Market: Browse and install skills from multiple providers (Aliyun, ClawHub, ModelScope) via a new Skill Market page with cross-provider search
  • feat: Adding eval for metrics analyses
  • Add tool safety docs
  • Fixes to http toolset
  • Extract codewhale-windows-x64.zip
  • Run install.bat (copies to %USERPROFILE%\bin)
  • Add %USERPROFILE%\bin to your PATH
  • fix: PyInstaller bundle is broken (missing agents SDK data + wrongly excluded gql); bump to 1.0.1
  • Agent: Adds a Browser component that enables AI to autonomously navigate and interact with web pages. #14888
  • RAG: RAPTOR construction now introduces AHC mode (Ψ-RAG), which expands semantics from the document level to the dataset level.
  • Agent: Introduces lightweight @tool decorator to streamline Python function registration process for chat models. #15047

May 26, 2026

  • MiniMax is now available as an OpenAI-compatible provider: configure MiniMax models through the OpenAI-compatible provider flow with primary and fallback endpoint validation, with macOS thinking…
  • Self-hosted assistants route runtime calls directly to their own ingress: when you connect a self-hosted assistant from the web app, message and event calls go straight to your assistant with the…
  • Billing and resize flows are clearer and safer: tier upgrades go through a read-only confirmation step before submitting to Stripe, plan resize triggers a confirmation modal after the Stripe…
Palmier Pro
  • add feedback entry to account popover footer
  • add in-app feedback window with optional screenshot
  • show discounted price in upgrade UI when backend coupon is active
  • feat: Better source-aware testing
  • feat: Migrate from Poetry to uv
  • Edit Strix GitHub Actions integration tip
  • Full Changelog
  • fix: Model handling for tool calling in agents and update IBM models
  • fix(ci): backport bundle rename step to nightly workflow on main
  • fix(ci): gate nightly release on successful tag creation
  • Hybrid AI. Mix and match local models (Ollama, LM Studio, etc.) with cloud providers (OpenAI, Anthropic, Google) in a single conversation. No manual switching!
  • You're in complete control. Create calculated rules that trigger on keywords, token counts, time of day, or image attachments instantaneously.
  • Save money without sacrificing quality. Route simple queries to cheap or local models. Reserve expensive API calls for the messages that actually need them.
  • bb desktop 0.0.10
  • Bedrock US: will use us-east-1
  • Bedrock Europe: will use eu-west-1
  • ae20f1f65175e4b1c2b9bbfc3f9b556a1589b7ec: feat: add aws bedrock europe (@andreynering)
  • Changes since langchain==1.3.1
  • chore(langchain): bump to 1.3.2, require langgraph>=1.2.2 fix(langchain): land final answer in last AIMessage for TodoListMiddleware feat(langchain): redact streamed PII in flight on PIIMiddleware…
  • chore(langgraph): bump version to 1.2.2
  • fix(langgraph): assign stable IDs to id=None BaseMessages before DeltaChannel checkpoint writes
  • release(checkpoint): 4.1.1
  • Project search now filters on the server — typing in the project picker sends a filter[name] query instead of filtering a local list, so results are accurate and instant even across hundreds of…
  • Results are paginated at 10 per page with a searching indicator — a visible loading state appears while results load, so the picker never feels frozen.
  • The entire session directory is zipped and uploaded — artifact uploads now capture everything in the session folder, not just individual files, so post-run inspection has the full context.
CodeWhale
  • Extract codewhale-windows-x64.zip
  • Run install.bat (copies to %USERPROFILE%\bin)
  • Add %USERPROFILE%\bin to your PATH
  • #1580 a1cd51b Thanks @threepointone! - Improve session auto-compaction estimates by including the Session-managed frozen system prompt, support custom token counters, and expose an auto-compaction…
  • #1559 f942ffe Thanks @cjol! - Stash chat turn recovery metadata before inference starts so interrupted pre-stream turns can be reconciled by chat recovery.
  • #1579 d0b4d0e Thanks @threepointone! - Ensure Agent-generated workflow instance IDs always satisfy the Workflows runtime ID validator.
  • (channels) add WeCom channel
  • (web) support externally-provided tools in Responses API
  • (gateway) add logs download button
  • Configurable skills cap — The maximum number of active skills is now configurable via max_active_skills in the skills plugin config and the settings UI, replacing the previous hard-coded limit.
  • Fix MCP multimodal content handling — MCP image, audio, and resource tool results are now properly preserved instead of being collapsed into empty text responses.
  • Improved browser iframe and shadow DOM support — A new browser DOM helper captures shadow DOM and iframe content with frame-chain/node references, improving the agent's ability to interact with…
  • ci: pin azure/setup-helm and helm/kind-action to commit SHAs
  • 04a6547bb019ef80c9f44dc1ec16659b88c3a0cf Document trustLoopbackProxy in config show output
  • 617c763e76dfce256ce1086097ebc72b5b0195d0 Merge pull request #1 from BH-M87/codex/add-trustloopbackproxy-to-config-show
  • 2f285bb6341e16b3ca01aa0b5d7a1d4ff00345fa Merge pull request #553 from pinchtab/fix/post-release-clawscan-hardening
  • npm install ERESOLVE on fresh install (PR #649, closes #631). @anthropic-ai/sdk bumped from ^0.39.0 to ^0.93.0 so claude-agent-sdk's peer is satisfied.
  • Non-OpenAI base URLs silent-404 (PR #649, closes #646, #628). buildChatUrl + buildEmbeddingUrl no longer blindly prepend /v1/.
  • CLAUDE_MEMORY_BRIDGE writes to a path Claude Code reads (PR #649, closes #625). Slug now preserves the leading - on POSIX absolute paths and drops the spurious /memory/ subdir, matching…
  • feat: support Realtime custom voice objects
  • fix: #3459 add opt-in recovery for missing function tools
  • fix: apply hardened http client default to MCP SSE transport
  • CONTRIBUTING.md at repo root (PR
  • OpenCode docs broken path fix (issue #172, reported
  • .continue variant SKILL.md synced from v2.34.0 to v2.43.0 (issue #159, 9 versions behind).
  • chore(memos-local-plugin): rename web/ to viewer/, drop unused site/ scaffolding
  • fix(memos-local-plugin): infer embedding dimensions for rebuilds
  • feat: enhance memos-local-plugin Windows installation experience
  • wren: add dbt import workflow (0bd16b8)
  • wren: build MDL manifests from OSI semantic models (983cf8b)
  • wren: honor SQL identifier case across policy, extract, and CTE rewriter (9b5fe22)
  • Pass session id to judge LLM calls
  • fix(agent): skip screenshots on new tab pages

May 25, 2026

  • feat(providers): derive discovery from profiles
  • docs: update NemoClaw/OpenClaw references
  • ci: seed shared Rust caches from main
  • 55721653ceb67f04abd53c278ade01d05043bf7d: feat(skills): add descriptions to skill picker and use attachements (@taciturnaxolotl)
  • 2b941aaa0d3b983e0c64e92b3b04a0e4ba855747: feat(ui): add ctrl+y keybinding to toggle yolo mode (@andreynering)
  • 092ac3e73a100d3ae7ec7fc16306de3fe5476f31: feat(ui): add scrollbar to sessions dialog (@andreynering)
  • MCP 作用域更准确:桌面端现在按 CLI 语义区分 local / project / user scope;创建或编辑 local/project MCP 时需要显式选择目标项目,避免写到当前 session worktree 或 server 启动目录。
  • MCP 设置页更可信:设置页会从活动项目、最近项目和已有 user-private MCP key 聚合配置;project-scoped 条目不会因为目标项目不同而消失,初次加载时也不会闪一下空列表。
  • 工作区文件引用更顺手:普通 dotfile / dotfolder 会在文件树中显示,文件和目录都可以从工作区菜单添加到聊天;常用复制路径改为项目相对路径,绝对路径保留为显式次级操作。
  • Modal sandbox provider is now a first-party plugin - Paperclip ships a Modal sandbox-provider plugin so companies can run agents on Modal's managed sandboxes alongside E2B, Cloudflare, and…
  • Workspace diffs are a first-class viewer plugin - The new workspace diff plugin renders staged, unstaged, head, renamed, binary, oversized, and untracked changes through host services and plugin…
  • Routines can carry their own secrets - Routine env now flows through the runtime contract with persisted revisions, agent < project < routine precedence, and safe secret metadata in routine…
AgentScope
  • chore(project): temporarily deprecate evaluate/module/rag/tts/realtime modules pending refactor
  • refactor(msg): simplify core building blocks and Msg class structure
  • feat(msg): formalize Msg type rules and constraints and add message related tests
  • Refreshed the conversation header on desktop: model selector moved to the right, agent logo sits next to the title editor, and team chats now show a Peoples icon.
  • Mobile experience is now noticeably more compact: tighter message bubbles (14px), shorter sendbox, ellipsis instead of marquee on the model pill, slimmer titlebar (no + button), and…
  • WebUI users now see the "Work in project" footnote on the new conversation page and can attach a workspace to fresh conversations, just like the desktop app.
  • Full Changelog
  • Load skills from ~/.agents/skills so they can be shared across agents
  • Restore the textual theme selection system with an onboarding theme picker
  • Surface unauthenticated connectors in /mcp with needs auth / needs setup labels and start the OAuth flow from inside the CLI
  • POSIX init-session.sh portability across the 8 mirrors (PR #169). The script's shebang is #!/usr/bin/env bash, but tests/test_init_session_slug.py:27 invokes it via sh, bypassing the shebang.
  • Install-scope transparency block in canonical SKILL.md (Turn-Loop Integration section).
  • Manual fallback procedure for /plan-goal and /plan-loop inline in the canonical SKILL.md.
  • Kill the entire process group on command cancel. Previously cancellation (client disconnect, timeout, DELETE /command) sent SIGKILL only to the bash group leader, so children spawned via & or…
  • Extend mitmproxy CA wait from 30s to 300s and log the actual wait duration. The bootstrap script previously gave up after 30s and started the sandbox without system CA trust, silently breaking…
  • @Pangjiping
  • feat: paste image bytes or file URL into media panel via ⌘V
  • fix: parent SwiftUI .onDrop shadowed inner drop targets on macOS 26
  • fix: panel dividers needed multi-grab because hit area was thinner than visible gap

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.