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 6 – Jul 12, 2026

198 stable releases from 81 tracked agents and frameworks in the week of Jul 6, 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.

Jul 12, 2026

  • Failures are investigated, not just reported — when a run fails, kane-cli automatically investigates whether it hit a product bug or a test issue and records a structured verdict in the pack's…
  • Result code 740 means a confirmed product bug — a confirmed bug is recorded as result code 740 in the run's result records and final event, so CI pipelines can distinguish genuine regressions from…
  • Proactive detection is configurable — --bug-detection off|stop|continue on run, testmd run, and testrun run (default off): stop halts the run on a confirmed bug, continue records it and keeps going.
  • #1550 f17a6ec Thanks @wbxl2000! - Treat a dismissed question prompt as the user choosing not to answer, instead of implicitly selecting the recommended option.
  • #1488 7bd29ab Thanks @starquakee! - Rename the dynamic tool loading model capability from select_tools to dynamically_loaded_tools.
  • #1497 c6e02da Thanks @sailist! - Add a print-mode background policy that lets kimi -p stay alive across background-task completions so the main agent can be steered into follow-up turns.
  • The .codex/hooks.json commands were POSIX only (sh, python3, 2>/dev/null, $HOME, a trailing || true).
  • Each hook now carries a per-hook commandWindows override, the mechanism OpenAI's hooks documentation sanctions. The POSIX command is untouched, so macOS and Linux are byte for byte unchanged.
  • On Windows you need Python reachable as py or python, and Git for Windows for the three shell hooks. See docs/codex.md.
  • New optional "Topic Index" for datasets — groups documents into topic clusters and a short overview to improve search context and answer quality.
  • Batch ingestion improvements — easier and faster bulk uploads with progress and resumable retries.
  • Search relevance and ranking upgrades — more accurate, less noisy results for common queries.
  • Factory: add API-key usage authentication with API-first Auto mode and recoverable fallback to the existing web session path. Thanks @araa47!
  • Developer tooling: add an offline adaptive-refresh replay CLI for comparing policy behavior against caller-supplied JSONL traces, without collecting production data. Thanks @hhh2210!
  • Settings: split provider pane "Settings" sections into "Menu bar" and "Connection" so metric pickers and auth/cookie/source controls are grouped by topic.
OpenSandbox
  • Credential Vault requires dns+nft enforcement. Credential-bound destinations can no longer be reached via direct-IP connections that bypass DNS.
  • X-Forwarded-Proto no longer trusted from arbitrary peers. Only trusted from IPs/CIDRs listed in OPENSANDBOX_EGRESS_CREDENTIAL_VAULT_TRUSTED_PROXY_CIDRS.
  • Match.Ports removed from Credential Vault bindings. Port is now derived from scheme (http→80, https→443) — the only ports the intercept layer redirects.

Jul 11, 2026

  • MCP server for CLI 3.0: shared MCP runner, max_dim guard.
  • Fix navigation readiness detection: per-target lifecycle event storage, and handle same-document navigations without burning the readiness timeout.
  • Fix markdown extraction destroying URLs and dropping long link lines.
  • Godot 4.2+
  • Active Ziva account
  • chore(deps): bump fantasy (0.35.2 -> 0.36.0) (@coloryourlife)
  • This release is a bug-fix patch: it plugs a WebAssembly memory leak, extends comment removal to more JavaScript/TypeScript module files, and fixes the token count tree for underscore-prefixed…
  • Parsed web-tree-sitter Tree objects were not released after use, leaking WebAssembly heap memory when parsing or compressing many files — noticeable on large repositories.
  • Special thanks to @isaka1022 for their first contribution! 🎉
AionUi
  • Fixed Windows builds shipping with an outdated Codex verification layout, which could cause Codex to fail verification after install.
  • fix(build): verify current Codex ACP Windows binaries
  • fix(build): align Codex installer verifier
  • Standalone deploy: CLM is wired into cube-sandbox-control.target; one-click and TencentCloud Terraform deployments both include it.
  • Redis service discovery: each CubeProxy replica registers its admin endpoint in Redis; CLM discovers live replicas and coordinates auto-pause / auto-resume with no static replica list.
  • Protocol compatibility: wire protocol, Redis schema, and SDK behavior are unchanged; existing AutoPause / AutoResume semantics are preserved.
  • [feat] elevenlabs Add voice isolator and dubbing workflows
  • fix: allow 65-point .cube LUTs
  • [feat] add audio scrubbing and timeline meter
  • Agent Sessions: opt in to discover, list, and focus live local or SSH-connected Codex and Claude Code sessions from the menu and CLI; discovery remains off by default.
  • Wayfinder: add opt-in local gateway health, routing, savings, and latency usage with configurable loopback URL support. Thanks @tcballard!
  • Menu bar: add a "Show reset time when quota runs out" option that replaces exhausted Percent, Pace, and Both values with the countdown until reset, then restores the selected metric afterward.
  • Affected: pydantic-ai / pydantic-ai-slim >= 2.0.0, < 2.5.0 (v2) and >= 1.88.0, < 1.107.1 (v1)
  • Patched: 2.5.0 (v2) and 1.107.1 (v1)
  • Not affected if every sensitive tool uses requires_approval=True / ApprovalRequiredToolset, or if your tool handlers validate their arguments and enforce authorization themselves.
  • feat: support GPT-5.6 request controls
  • feat: add hosted multi-agent beta support
  • fix(sandbox): own Daytona PTY worker tasks
  • Updated bundled Claude CLI to version 2.1.207
  • Fixed CI workspace trust so Claude Code honors project-scoped permission grants in checkout directories
  • feat: update version to 1.0.2
  • docs: update action count to 10,000+
  • feat: add MCP connection configuration and update ResourcesPage layout
  • fix: add configurable provider retry attempts
  • fix: add missing OPENAI_COMPATIBLE_API_KEY to secret redaction
  • fix: constrain init update writes to openwiki
  • Rename the packaged skill identity back to browser-harness.
  • More robust CLI telemetry: skip the daemon ping when telemetry is off, log browser type (remote/local).
  • Suggest the cloud browser when no local browser is available.

Jul 10, 2026

Omnigent
  • 🖥️ On older desktop builds without the embedded browser, the Browser tab is now hidden instead of showing a dead tab that does nothing
  • Changes since langchain==1.3.12
  • release(langchain): 1.3.13 feat(langchain): add meta extra and support langchain-meta in init_chat_model feat(openai): support explicit prompt caching chore(deps): refresh lockfiles
  • The one where you pick a default AI provider once and forget about it, your images and attachments work properly, and your assistant's mobile app gets a full design polish.
  • Pick a default provider and stop thinking about it. Mark one provider connection as "default" and your assistant always tries that one first.
  • Errors that make sense. When a model call fails, you get a plain reason like "rate limited" or "invalid API key" instead of a status code you have to Google.
Supermemory
  • Pluggable embeddings — local ONNX (default) or OpenAI / OpenAI-compatible / Google via SUPERMEMORY_EMBEDDING_*; first-boot picker + plan lock in embedding-plan.json
  • Safe upgrades — populated stores without a lock assume legacy local · Xenova/bge-base-en-v1.5 · 768d; same-dimension model switches fail fast instead of mixing vectors
  • GPT-5.6 模型支持:OpenAI 官方登录新增 GPT-5.6、GPT-5.6 Pro 和 GPT-5.6 Codex 模型目录、选择和迁移支持,并补齐模型元数据与会话恢复链路。
  • 长任务恢复更可靠:长时间运行的 turn 在流式传输或 WebSocket 中断后能更准确地区分可恢复与已完成状态,避免界面长期停留在运行中或丢失最终 assistant 输出。
  • 权限交互更安全:同一会话中的并发权限请求会分别保留和响应;活跃 turn 期间禁止切换权限模式,避免运行时权限状态与界面选择不一致。
  • feat(desktop): default observer archive on for dev-nest builds (d9e4edbbf)
  • feat(desktop): thread baked build env into global agent config UI (2f2ad409a)
  • feat: relay invite links (mint + claim + landing page + deep link) (2e529aab7)
  • Add bash permission rules engine with command parser, allow/ask/deny rules persisted in .dexter/settings.json, and an "always allow" approval option for simple commands
  • Add GPT 5.6 model family (Sol, Terra, Luna) as the default OpenAI models, with auto-upgrade from gpt-5.5/5.4/5.2
  • Add insider ownership tool for SEC Forms 3/5 insider holdings (positions, options, RSUs)
Strands Agents
  • feat: expose client_name parameter on MCPClient for clientInfo identification (612f47f7)
  • feat(port): allow ports in either direction (e385f74a)
  • feat: added span redaction (2d6e6502)
  • feat(app-server): enrich final archive manifests and remove initial snapshots
  • feat(enterprise): make BYOR key alias pattern configurable
  • fix(frontend): mention SMTP env vars for budget alerts
  • Crash-safe chat persistence — Serialized chats are now written to a temporary file, fsynced, and atomically swapped into place, so an interrupted save never corrupts existing chat data.
  • Circuit breaker for runaway response loops — A configurable limit on consecutive malformed or repeated model outputs prevents infinite loops.
  • Gemini CLI support in Orchestrator — Gemini CLI is now a registered orchestrator backend with auth detection, headless workflow support, and exposed settings and status metadata.
  • api_server modularization completed — 1,103 → 371 lines (#424 closing #331,
  • Centralized env config — one Pydantic EnvConfig schema + an AST CI gate that rejects raw os.getenv outside the config layer
  • Trading 212 read-only connector
  • #1542 f80b2ea Thanks @wbxl2000! - web: Fix the "Turn finished" desktop notification and completion sound firing twice per turn.
  • #1535 04041eb Thanks @wbxl2000! - web: Hide the internal image-compression note so it no longer renders as user message text.
  • #1536 db61c9e Thanks @RealKai42! - Stop unsupported image formats (AVIF, BMP, TIFF, ICO, …) from breaking sessions at every entry point — including remote image URLs and images mislabeled by a…
  • correct Vertex AI legacy model ids by removing stray asterisk (60a9794), closes
  • deps: update k8s.io/utils digest to be93311 (de01885)
  • deps: update module golang.org/x/net to v0.55.0 [security] (2e57a6f)
  • Runtime 2.0: Refactor the kernel based on AgentScope 2.0
  • Agent OS Driver: Unified management of MCP/A2A/ACP connections, encrypted credential storage, and per-call approval policy
  • Loop Engineering: New Loop gate design with three Loop templates: Normal Mode / Mission Mode / Goal Mode
  • sync model reasoning options.
  • package Claude sidecar for target architecture.
  • fix(agent-gui): sync model reasoning options
  • fe46dfdbf66599d20a584d5e720e14ea4e4f8ac4 MUL-4203: Fix Cursor MCP auth source seeding (ZIC-52)
  • f599333f9728dc68f14439fadc0481735e00499f [MUL-4348] Route coalesced replies per root thread
  • 37b0a6e72e45b0336d7d444485f6fd39165cc9a9 docs(changelog): add v0.3.43 release entry (MUL-4376)
WrenAI
  • add wren-core-wasm module with browser WASM support (5165a20)
  • core: add BigQuery EXTRACT function support and window frame validation (eae6a34)
  • core: apply default nulls last policy for ordering (2477f43)
  • Dart and C++ test/benchmark ignore patterns —
  • Scala (#553, from #493) and Swift —
  • Modern TS type-declaration extraction fixes —
  • agent: Added GPT 5.6 Sol & Terra for ChatGPT subscription. Note: GPT 5.6 Luna is not available yet, since OpenAI has not unlocked access for third-party clients
  • feat(llm): add built-in Eden AI provider preset
  • fix(llm): make OCR_LLM_EXTRA_HEADERS apply to all resolver strategies
  • fix(vscode): add PNG icon for marketplace display and bump to v0.1.1
  • Godot 4.2+
  • Active Ziva account
  • feat(council): support model fallback chain per councillor
  • fix(hooks): restore ephemeral post-file-tool-nudge to prevent duplicate PHASE_REMINDER
  • fix(scheduler): reconcile running jobs when child session goes idle
  • 0.6.4 changelog
  • [telemetry] analytics
  • Validate video-to-audio span in agent path; shrink scoring renders
  • Fixed the crash (SIGSEGV) when opening any plugin page — see the note above (also the likely cause of the plugin crashes reported in #159)
  • Incompatible plugins are now rejected with a clear "update this plugin" message instead of taking the app down
  • Plugins carry their compiler fingerprint so a mismatch can never silently pass again
  • Full Changelog
  • Added the --agent flag and agent/agents subcommand, allowing users to select a custom agent at launch and list available agents.
  • Added in-file keyword search (/) and jump navigation (n/N) to the artifact detail viewer, allowing users to find and cycle through matches without disrupting terminal escape sequences or image grids.
  • Fixed print mode (--print / -p) silently exiting with a success code and empty output when a request failed server-side, now writing the error to stderr and returning a non-zero exit code.
  • agent-framework-core: Add message injection middleware so tools or host code can enqueue messages into an active run and drain them into the next model call within the same AgentSession
  • agent-framework-core: Integrate message injection into create_harness_agent and the harness console sample so a running harness agent can be nudged mid-turn
  • agent-framework-core: Add progressive MCP disclosure so agents can discover, load, and unload MCP tool schemas on demand while keeping the allowed_tools boundary intact
  • ebd845c003579484171922fda8a671367eee7684: feat(agent): send session hash in header for cache affinity (@meowgorithm)
  • 5b61432984ba271cf9cb3791295ba94e9c2c8f81: fix(release): drop broken .termux.deb artifact (@Ricardo-M-L)
  • e3074f14cec703bdddba13e6df5cf2242b34194c: fix(server): show update notices in client-server mode (@meowgorithm)
  • release(langgraph): 1.2.9
  • fix: updateState metadata/counters for delta channel

Jul 9, 2026

  • Updated bundled Claude CLI to version 2.1.206
  • feat: add GPT-5.6 model defaults and migrate examples
  • fix: support cache-write usage across OpenAI Python versions
  • fix(realtime): make session cleanup deterministic
Pi
  • max thinking level - New opt-in thinking level above xhigh, natively supported on GPT-5.6 and adaptive Claude models, available across CLI (--thinking max), SDK, RPC, and model selection.
  • Input-based pricing tiers - Request-wide input-token pricing tiers for accurate long-context cost accounting (e.g.
  • Added the opt-in max thinking level across CLI, SDK, RPC, model selection, and themes. Custom themes can define thinkingMax; existing themes fall back to thinkingXhigh.
OfficeCLI
  • Full Changelog
  • Full Changelog
OpenHands
  • chore(main): release 1.11.0
  • open_ai: Added support for GPT 5.6 Sol/Terra/Luna
  • cli: Fixed an issue where the previous workspace would not be restored when using cli_default_open_behavior: new_window and no path was provided
  • fix(google): broaden _is_gemini_3_flash_model to cover all Gemini 3 Flash variants
  • fix(assemblyai): respect mode preset for turn-silence; remap deprecated u3-pro
  • Move frame processor url/token/stream info to client sdk
OpenWiki
  • chore: bump langchain-openai dependency
  • docs: update OpenWiki
  • chore: bump langchain openai to version 1.5.5
OpenCode
  • Prevent crashes and bad pricing data when GitHub Copilot returns models with a zero billing batch size.
  • Add a model-specific system prompt for Meta Muse Spark.
  • bb desktop 0.0.29
Tutti
  • Fixed an issue where the agent failed to load the model catalog via relative paths in the sandbox, ensuring correct model availability.
  • Improved model configuration synchronization during agent startup to reduce initialization delays caused by path resolution errors.
  • fix(agent): mirror relative Codex model_catalog_json into sandbox
  • Fixed URL intent routing so custom Agent Zero schemes are no longer misrouted to the browser handler
  • Replaced the 50 ms idle poll with a blocking frame read so live screencast delivery follows the producer cadence
  • Negotiated binary and slim screencast frames to avoid base64-in-JSON overhead and repeated metadata on every frame; clamped capture to subscriber viewport and DPR with lower streaming JPEG quality
  • #1501 b91099e Thanks @liruifengv! - Display Extra Usage (fuel pack) balance in /usage and /status commands.
  • #1517 173bdfd Thanks @chengluyu! - Fix resuming sessions whose original working directory no longer exists.
  • #1516 9fb1915 Thanks @chengluyu! - Fix prompt-mode goals so they run until completion and report invalid goal commands before sending prompts.
  • fix(driver-podman): gate Linux-only Path import
  • docs: fix Docker version format from 28.04 to 28.0
  • docs: update man page date to 2026
  • fix(dashboard): price proxy savings without litellm
  • fix(proxy): surface codex websocket loop failures in livez
  • fix(savings): guard non-finite numeric coercion
  • feat: add Python code review rules
  • feat(action): extract reusable composite PR-review GitHub Action
  • feat: enhance token usage resolution for OpenAI and Anthropic compati…
  • Experimental managed bash tool for running and tracking shell sessions
  • disabled-tools CLI flag to turn off specific tools for a run
  • Per-tool description overrides via <tools-dir>/prompts/<name>.md
  • OAuth on the AgentOS MCP Endpoint: Added OAuth support for the AgentOS MCP endpoint via AgentOS(mcp_auth=...). See cookbook.
  • AG-UI Client Tools: Added client_tools support for AG-UI frontend tools.
  • agno connect Enhancements: Multi-target select, disconnect, restart hints, and identity-named entries.
  • 技能市场上线:新增桌面端技能市场,聚合 ClawHub 与 SkillHub,支持搜索、筛选、分页、详情、文件预览、安装确认、卸载和已安装状态识别。
  • 技能安装更保守:市场安装链路补齐风险标签、扫描摘要、来源状态、免责声明和不可确认风险阻断,避免把上游不完整或高风险结果误展示成安全可装。
  • 会话活动面板:新增右侧活动面板,集中展示 tasks、background tasks、SubAgent、team activity 和 sources;SubAgent 运行可进入独立详情页查看。
  • f30898f3ec80e58f7635272069002c72ea5659ea MUL-4299: guard migration numbering
  • 01f28e8af6282c62d7459f757dd8c68b614fd718 docs(changelog): add v0.3.42 release entry across en/zh/ja/ko
  • f8c4c881293c146ae5a160f7ba116195f9998355 feat(agents): show runtime model + effort on agent hover card [MUL-4280]
  • Fix #1540: fix
  • Fix #1888: [Bug] test_system_parser.py: SystemParser.init() got an unexpected keyword a
  • Fix #1525: [Bug] clean_json_response crashes with cryptic AttributeError when given None
  • feat(mcp): add server instructions
  • feat(mcdonalds_cn): add McDonald's China provider
  • feat(deployment): add Fly.io deployment configuration and docs

Jul 8, 2026

  • Changes since langchain==1.3.11
  • release(langchain): 1.3.12 fix(langchain): propagate interrupts through ToolRetryMiddleware fix(langchain): avoid shared process-group kill in shell middleware fix(langchain): sanitize anthropic…
Claude Agent SDK
  • Updated bundled Claude CLI to version 2.1.205
  • d0dc9fc9dfce76b9e35b2686ee4bfccfca796222: feat: prepare for gpt-5.6 (@andreynering)
  • cb1292026cc113e2a2e61f14686f2a36551f74a0: fix(bangmode): don't add extra ! when browsing history (@meowgorithm)
  • 72c5668eface73b6e3745cccc8098f2cd1e55879: system prompt: avoid emdashes in source code (@meowgorithm)
OpenHands
  • feat: rename admin dashboard to usage & monitoring
  • feat: add SMTP email service
  • feat: track user login timestamps
  • chore: Fast fail, dont fallback to diff model
  • fix: validate configured provider before setup completes
  • feat: use openai responses api
  • bb desktop 0.0.28
  • Agent: Added setting to exclude project rules files from commit message generation prompts (agent.commit_message_include_project_rules). (#56867; thanks ilavisharma)
  • Added llama.cpp as a language model provider. (#59964; thanks hanouticelina)
  • Agent: Changed the default behavior of queued messages for the Zed agent so that they get send at the end of the generation.
OfficeCLI
  • Full Changelog
  • #1372 4a3100b Thanks @omeraplak! - Expose AI SDK usage semantics in agent onEnd outputs: usage now represents final-step usage, totalUsage represents aggregate usage across all steps, and steps is…
  • docs: fix STYLEGUIDE heading to match filename
  • docs(kubernetes): bump cert-manager to v1.20.3
  • docs: fix article before OpenShell in sync-files
Kimi Code
  • #1506 e83511a Thanks @7Sageer! - Fix a misleading "OAuth login expired" message shown when a model is not available for the current account; the CLI now shows the provider's actual error.
  • Heartbeats were renamed to schedules: mastra.heartbeats → mastra.schedules, config heartbeat → schedules, type renames (e.g., Heartbeat → AgentSchedule), new agent schedule IDs use agent_, and the…
  • Server removed /api/heartbeats/* routes in favor of the unified /api/schedules/* API.
  • Client deprecated *Heartbeat() methods in favor of unified schedule methods (to be removed in a future release).
  • PyPI: bash pip install crawl4ai==0.9.1
  • Docker: bash docker pull unclecode/crawl4ai:0.9.1 docker pull unclecode/crawl4ai:latest
  • Note: Docker images are being built and will be available shortly. Check the Docker Release workflow for build status.
  • feat(ts): preserve MCP output schemas (b07fadb0)
  • feat(designs-storage): add storage design doc proposal (2e3dfff1)
  • feat(memory): pass per-message sequence numbers to add_messages (e050fc8b)
  • chore(deps): bump mcp to 1.28.1 to fix 3 high-severity Dependabot alerts (14439528f)
  • fix(SKY-11997,SKY-12004): name session-scoped downloads by configured download_suffix
  • fix(SKY-11948): consume output-contract grants at dispatch
  • chore(docs): update WeChat group QR code to wx-15
  • feat(settings): assistant editor and settings UI polish
  • feat(feedback): attach core diagnostics to reports
  • DocumentNDCGEvaluator now matches documents by their content field by default instead of their auto-generated id.
  • Added native asynchronous support (run_async) to LLMEvaluator, FaithfulnessEvaluator, and ContextRelevanceEvaluator.
  • Added optional YAML frontmatter extraction to MarkdownToDocument. When initialized with extract_frontmatter=True, YAML frontmatter at the beginning of a Markdown file is removed from the converted…
  • b32e83916c407152d7d108235a05e77231d660e8 Merge branch 'main' into dependabot/go_modules/golang.org/x/net-0.55.0
  • db21255ca81619383b3505f7432ad17cb7295121 Merge pull request #605 from pinchtab/fix/issue-602-npm-postinstall
  • 28f1b56c9b08700d93c384dc9bbabe717d437050 Merge pull request #606 from pinchtab/dependabot/go_modules/golang.org/x/net-0.55.0
  • feat: release 0.86.2 @MrChaker
  • feat: add endpoint to export queues to promethues @MrChaker
  • fix(builder): surface internal error cause in run-details step panel @Jaimin791
  • add codex managed node fallback.
  • fix(agent): add codex managed node fallback
  • macOS Apple Silicon (arm64)
  • 947a54b67496231576d12f6e625805a995615d52 docs(agent): align state guidance with sweep
  • c05b67ae4a9f43170394da2a50e68cd7ccd51f80 docs(changelog): add v0.3.41 release notes (en/zh/ja/ko)
  • 519d2aeff073ec6989f0966d018f573e64cc4174 feat(cli): enable issue ordering via cli (issue reorder, --position, --sort/--direction)
Open Connector
  • feat(headers): add caching headers for assets and catalog apps
  • feat: enhance overview page with call trend and recent calls sections
  • Out of the box: the full Agent runtime is bundled — launch right after install, no need to set up Python or other dependencies
  • Full chat experience: streaming replies, session management, tool-call step display, Markdown rendering, plus sending and previewing images / videos / files
  • Visual management: configuration, models, knowledge base, scheduled tasks, skills, and memory pages mirror the Web console, all manageable in the native UI
  • cuVS GPU 向量检索后端:新增 storage.vectordb.backend = "cuvs",支持 cuVS brute-force 精确检索和 CAGRA 近似检索;也可以在 backend = "local" 下启用 cuvs.auto_enable,让系统在 GPU 内存足够时自动使用 cuVS,否则回退到原生 CPU 索引。
  • 递归网页导入:普通网页导入改为 Scrapy 驱动的同站 BFS 爬取,支持 depth、max_pages、include_paths、exclude_paths、allow_external_links、skip_download_links;正文抽取切换到 trafilatura,并保留 sitemap/RSS 的整站导入路径。
  • 记忆 v3 与训练框架:OpenViking 现在始终使用 v3 记忆抽取链路;新增流式记忆更新器,将并发提交按 peer 和 memory type 分组合并,再用 patch-merge 应用。cases、trajectories、experiences 可以参与 session train 流程,Tau2 train/eval 也有专门入口。
  • On-device speech analysis — Detect speech, dead air, and speakers across clips. Remove silence from the timeline or via the agent. New Audio tab with Speech controls.
  • On-device beat detection — Detect beats and downbeats on audio clips (clip menu or agent). Snap drags, trims, and drops to the beat grid with waveform ticks.
  • Better clip sync — Sync clips by timecode, audio, or auto (timecode first, then audio). Improved audio sync with capture-date seeding.
OpenSandbox
  • Added directory listing support through the new execd directory API. The SDK exposes depth-controlled traversal (depth defaults to 1 for immediate children) and a FileInfo.type field…
  • Added line-based file reads through GET /files/download with offset (1-based line number) and limit (line count) query parameters, enabling callers to read large text files by line range.
  • Added detailed replace-content feedback without breaking existing callers. The replace API now returns ContentReplaceResult[] exposing per-file replacedCount, so callers can detect no-match (0)…
  • feat(viewer): dark mode + system monospace font
  • Feat/telemetry http exporter
  • Agent execution mode cycling is now publicly available: default -> accept-edits -> plan)
  • Added request-review (default) mode as the default execution behavior: automatically pauses before file write operations to display an interactive, line-level diff preview (f shortcut) where users…
  • Added an Agent Mode option to the /settings panel so users can set and persist a default execution mode (default, accept-edits, plan) without manually editing settings.json or passing --mode on…
Hermes Agent
  • fix(whatsapp): unpin Baileys from git commit, use published 7.0.0-rc13 — the WhatsApp bridge dependency now installs from the published npm release instead of a pinned git commit, making installs…
  • Pull latest LLM models dynamically in the crew wizard.
  • Support inline skill definitions.
  • Add generated Flow Definition authoring skill.

Jul 7, 2026

Herdr
  • The session navigator now keeps the active search query when leaving and re-entering search focus, and its footer now shows shortcuts for the current input mode.
  • Re-focusing an already-focused done agent or pane through the socket API now marks it seen instead of leaving stale done status in API responses.
  • Windows foreground-process detection now ignores cyclic process-parent snapshots instead of growing memory until the server aborts.
  • fix(mcp): route empty-API-key 401s to the Configure dialog instead of OAuth (bee2b836c)
  • fix(SKY-12021): show user-inputted prompt for file download block (323adca6a)
  • fix(SKY-11991): expand collapsed loop/conditional ancestors before block jump (bff497e2e)
OpenHands
  • chore(main): release cloud 1.42.0
Agno
  • agno 2.7.1 + agnoctl 0.1.1.
  • A better onboarding. New users now answer a few questions about their work, preferences, and the personality they want before meeting their assistant.
  • Work and personal calendars, together. Connect work and personal accounts (Google or Outlook).
  • Channel permissions got stricter. Permissions for channels (Slack, Telegram, etc.) are now checked every single time your assistant tries to use a tool, not just once when the channel connects.
  • interactions: support for GenAI SDK upgraded to 2.9+ (7a9152a)
  • cbb6daaa80841614abc091f3201c5e635895c22e: feat(ui): preserve newlines in expanded tool content (@taciturnaxolotl)
  • d341d84b24fbe6471708c15f4dc08533a9ea6dc4: fix(alibaba): fix missing thinking traces on deepseek via aliababa (@andreynering)
  • 3ac7f1c48bbe3031772fb2f47c2f356283790591: fix(baseten): make the thinking on/off toggle work for baseten (@andreynering)
  • Updated bundled Claude CLI to version 2.1.203
  • feat(ts): preserve MCP output schemas (b07fadb0)
  • feat(designs-storage): add storage design doc proposal (2e3dfff1)
  • feat(memory): pass per-message sequence numbers to add_messages (e050fc8b)
  • feat: add global variable model overrides
  • fix: use session exec in telemetry writer
  • fix: disable mem0 telemetry by default
  • feat: add an OpenAI-compatible provider with a required base URL
  • fix: blank LangSmith input should disable tracing
  • fix(update): skip agent run when no repository changes are detected
  • fix(desktop): preserve agent model/provider when persona snapshot fields are blank (a3ee2c569)
  • feat(acp,desktop): identify and reap stale agent harness processes (3c6c0d447)
  • feat(desktop): active-draft badge, send-from-drafts confirm dialog, thread-deleted state (b394e95ef)
Tutti
  • Fixed an issue where minimized browser nodes remained visible, cluttering the interface.
  • Resolved incorrect rendering of webviews when minimized, ensuring proper window state management.
  • Enhanced visibility control for browser tabs within the workspace for a cleaner look.
  • docs: fix STYLEGUIDE heading to match filename
  • docs(kubernetes): bump cert-manager to v1.20.3
  • docs: fix article before OpenShell in sync-files
  • Better classify Z.ai context-window overflow errors so oversized requests surface the right failure mode (@fengjikui)
  • Handle unavailable config directories more gracefully when reading config files
  • Restore model details tooltips in the model picker (@arvsrn)
Semantic Kernel
  • 35ba23e1b3092271c778ca057afe1a796e16e70e .Net: Update package version to 1.78.0
  • e6c9673684ca03621885083faff1644e1f42695e .Net: Disable automatic HTTP redirects in HttpPlugin and WebFileDownloadPlugin default clients
  • f25753be0a126138d2eea39b0ca2985252dde25a Bump Scriban from 7.2.0 to 7.2.5 to fix NU1902 vulnerability
  • User-specific runtime secrets - Secrets can now be scoped to the individual human operator, not just the company.
  • Work Timeline - A new company-scoped Work Timeline page renders a compact, Gantt-style SVG view of when your agents worked, how handoffs happened, and where work overlapped — so you can scan…
  • Custom sandbox images with a built-in SSH terminal - Build reusable custom sandbox images and set them up without ever leaving the browser: an embedded SSH terminal lets you install packages and…
  • NLP/Tokenization: Adds a language-aware Snowball stemmer supporting 16 languages, integrates the dataset language parameter across the tokenization pipeline, and adds Dutch to the frontend. #14140
  • The system crashed with a ValueError when parsing LM-Studio model names containing an '@' symbol. #16467
  • The MCP server crashed because the list_chats function expected a list from the /chats API but received a paginated dictionary instead. #16639
Open Code Review
  • fix(tool): resolve file_read paths against git top-level in monorepos
  • Support agent interruption with context handling #1995
  • Integrate ReMe based longterm memory #1972
  • Integrate agentic longterm memory #1927
  • Added the LoCoMo benchmark runner under benchmarks/, including TOML configuration, an environment template, staged ingestion/search/answer/judge execution, result artifacts, and benchmark…
  • Improved hybrid search with heap-driven lazy expansion and global top-N competition. The 1.1.1 benchmark package reports stable LoCoMo targets around 91% for hybrid search and 93% for agentic search.
  • Synced the 1.1.1 source package while preserving GitHub-only workflows, templates, contributor-doc checks, public release support, use-case material, and the existing English README.md structure.
  • 274741638060fb34e524a5290e55bf38e15a9092 MUL-4117: feat(cli): add workspace member invite command
  • 33bd8aeaa97dcb2a625b5fc29a869f455462f634 MUL-4134: fix(lark): preserve same-agent bindings when reconnecting a revoked Feishu bot
  • 1de0c7d14ca58be457e274a87834f61edf41e353 MUL-4158: allow deleting orphaned profile runtimes
BrowserSkill
  • fix(extension): never return a borrowed tab into another session's Agent Window
  • fix(daemon): reject cross-browser session events from non-owners
  • fix(extension): don't fail-open the borrow gate on a transient tabs.get error
  • 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 codewhale-windows-x64.zip
  • Run install.bat (copies to %USERPROFILE%\bin)
  • @mastra/playground-ui removed root named exports; import from explicit subpaths (e.g. @mastra/playground-ui/components/Button).
  • @mastra/playground-ui removed Searchbar; use InputGroup composition instead.
  • Added opt-in storage retention. Declare per-table maxAge policies in the retention config, then call storage.prune() to delete rows older than their age.
  • First public release of FinRobot Desktop — an open-source desktop research & valuation cockpit for financial analysts (macOS, Apple Silicon).
  • Install: download FinRobot_0.1.0_aarch64.dmg below and drag FinRobot into Applications.
  • First launch: the app is not yet Apple-notarized, so macOS will report the downloaded app as "damaged". Copy and paste the following command into Terminal, then open FinRobot again
  • npm: bash npm install -g base44@latest
  • Homebrew: bash brew install base44/tap/base44
  • Version: 0.1.2 NPM Tag: latest
  • #1432 25a655c Thanks @RealKai42! - Preserve prior turns' thinking by default on the Anthropic provider (Claude and Kimi's Anthropic-compatible mode), matching the Kimi default.
  • #1451 16dc940 Thanks @wbxl2000! - web: Recover chat streaming after a stale background-tab WebSocket instead of requiring a page refresh.
  • #1456 e9ef939 Thanks @chengluyu! - Fix goal completion and blocked updates to produce one final user-facing outcome summary from the tool result.
  • feat: update default realtime model to gpt-realtime-2.1
  • feat: add Unicode storage option to SQLAlchemySession
  • fix(visualization): keep fillcolor on handoff nodes by merging style attributes

Jul 6, 2026

  • CLI: add responsive codexbar cards and compact --brief terminal usage views. Thanks @DonnieFi!
  • Antigravity: show pace details for legacy model-family and current session/weekly quota rows without changing compact icon lane semantics. Thanks @Zihao-Qi!
  • Widgets: make Kimi available with Weekly, Rate Limit, and Monthly quota rows. Thanks @joeVenner!
  • chore(release): patch v0.1.7 — post-release fixes + model-adaptive effort selector (0f5d970)
  • chore(release): stamp v0.1.7 [skip ci] (21c1e6b)
  • chore(release): v0.1.7 — deep-scan hardening pass + ultracode (a6a092e)
  • feat: add invalid final output recovery handler
  • fix: propagate realtime session iterator cancellation
  • fix: redact realtime validation failure logs
Buzz
  • chore(deps): bump crossbeam-epoch to 0.9.20 for RUSTSEC-2026-0204 (75b52ad7e)
  • fix(archive): atomic remove-kind path + split test modules (484bede75)
  • feat(archive): add agent turn-metric (kind 44200) local archive (429f54b4f)
  • Zombie CLI subprocess prevention: Shielded subprocess cleanup from asyncio cancellation so SIGTERM/SIGKILL teardown always runs, preventing orphaned claude child processes when the parent task is…
  • Silent whitespace loss on large NDJSON lines: Fixed the NDJSON parser silently dropping whitespace when a single line exceeded the 64 KiB stream buffer, which could corrupt tool output or…
  • TypeError on non-dict message content: Fixed an uncaught TypeError when the CLI emits a message whose content field is a plain string or other non-dict value instead of the expected list of…
Open Interpreter
No release notes published.
  • release(langgraph): 1.2.8
  • fix: delta channel bug with updateState on fresh thread will force snapshot instead of stub checkpoint
  • chore(deps): bump the minor-and-patch group in /libs/langgraph with 8 updates
  • One connection for all Vellum-hosted models. Before, if you used the models Vellum hosts for you, each one needed its own connection in settings. Now there's just one.
  • Check your credit balance from the terminal. Your assistant can run assistant platform credits to see how much you have left.
  • Control who can talk to your assistant. Each channel (Slack, Telegram, Phone) now has its own settings for who can reach your assistant and what they're allowed to do.
  • Added a code mode MCP adapter for running confined orchestration scripts against connected MCP tools.
  • Hid the execute tool unless code mode is enabled.
  • Fixed paginated MCP tool catalogs losing tool metadata and output schema validation.
  • fix(pptx/xlsx/docx): notify watch SSE after resident batch edits
  • @sixtycat2000-ctrl made their first contribution in
  • fix(config): read config files as UTF-8 regardless of locale
  • fix(models): don't mask malformed tool-call errors as truncation
  • docs: add Hermes Agent to supported agents table
  • rfc-0006: add driver config passthrough proposal
  • chore(deps): bump docker/login-action from 4.2.0 to 4.4.0
  • #1417 79b360c Thanks @RealKai42! - Enable Preserved Thinking by default for Kimi models when Thinking is on, keeping prior reasoning across turns.
  • #1073 6c0ce09 Thanks @sailist! - Add server APIs to restore archived sessions and list only archived sessions.
  • #1369 f0896a5 Thanks @starquakee! - Add experimental progressive tool disclosure (select_tools).
OpenHands
  • feat: implement semantic file chunking using tree-sitter AST parsing
  • feat(org): Add organization conversation admin dashboard
  • feat(app-server): capture production workspace state — initial snapshot at start + archive before delete (APP-2403)
  • Assistants moved out of Settings into their own top-level sidebar entry (above Scheduled Tasks).
  • Official built-in assistants are now off by default instead of flooding every selection list, so your own CLI assistants aren't buried. (Existing setups are migrated automatically.)
  • Opening Settings now lands on the Agents tab. Skills and tools are split into separate entries, page headers follow one consistent paradigm, and the agent availability filter…
  • feat: release 0.86.1 @MrChaker
  • fix(chat): prefer inline expressions over CODE steps in flow building @utopianguide
  • fix(flow-run): serve a POST confirmation page for approval resume links @othmanemad
  • feat(pages): add per-chapter routes for docs
  • fix(tui): persist official-tab models and refine saved secret hint
  • fix(pages): adjust HeroSection terminal section height and bottom padding
  • 5dfb0bec0690cc6a5b4c782d98ea93a05c3d7b03 Fix Codex MCP allowlist config rendering
  • ebd0248be25a1675bce29b580b046454074c344e MUL-4016: fix mention tokenizer stacktrace backtracking
  • 083e045ec617b6c6117eed650207e98be24090d2 MUL-4103: harden Windows browser MCP config
  • Fix 1.x compatibility with ACP: Pin ACP version to resolve a breaking change in ACP that caused historical 1.x versions of QwenPaw to malfunction. (@rayrayraykk in
Tutti
  • Fixed issues with session activation failures, enhancing workbench startup reliability.
  • Corrected permission validation logic in background service task submission to prevent interruptions.
  • Internal updates to session status monitoring improve data migration safety during upgrades.
  • Connectors
  • Add BigQuery target connector
  • feat: add Azure Blob source connector
Planning With Files
  • Python suite: 200 passed, 5 skipped, 0 failed (12 new opt-out tests)
  • Functional smoke test on a live temp plan: with the variable set, no hook output, tool calls still allowed, plan files byte-for-byte unchanged
  • scripts/sync-ide-folders.py --verify: all IDE folders in sync
  • memory: add 'wren memory watch' to auto-reindex on source changes (d5911c5)
  • policy: govern data-reading TVFs in all positions + structure source-func allowlist (5ff51e4)
  • wren: add cross-dialect type translation to type_mapping (e124ff5)
  • MCP path (~2 min): claude mcp add commonly … -- npx -y @commonlyai/mcp connects Claude Code / Cursor / Codex with the full 22-tool kernel surface (@commonlyai/mcp@0.1.7).
  • Autonomous path: npm i -g @commonlyai/cli → commonly agent attach claude|codex → commonly agent run — your local CLI becomes a pod member that answers @mentions and DMs while you're away, with…
  • Agents ship with house rules built in (the bundled behavior skill): they orient before posting, read attached files before answering, react instead of posting "got it", reply-thread to the right…

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.