OpenClaw Newsletter - 2026-04-23
OpenClaw Newsletter
Thursday, April 23, 2026Top Stories
- fix: preserve nvidia-prefixed model refs in session overrides and config writes — Summary Fix NVIDIA model-id handling for Nemotron refs that include an upstream nvidia/ namespace. This PR fixes two related paths: 1. Persisted session / fallback override normalization no longer...
- fix(loop-detection): escalate generic_repeat to critical at criticalThreshold — Problem The generic_repeat detector in tool-loop-detection.ts only ever returns level: "warning", meaning it can never block execution — even when a tool has been called 20+ times with identical...
- Measure Telegram QA reply RTT — Summary - record per-scenario Telegram QA reply RTT in reports and JSON summaries - cache Telegram forum metadata lookups so group command handling avoids repeated getChat calls Validation - pnpm...
- fix(ui): remove duplicate section headers in single-section settings view — Summary - Problem: In the Control UI dashboard settings, when opening a single section (for example, Automation), the section title and description are shown twice: once in the top section hero and...
- fix(device-pairing): coerce array-typed state files to plain objects in loadState — Summary - Problem: loadState() in src/infra/device-pairing.ts:111 and src/infra/node-pairing.ts:141 uses nullish coalescing (pending ?? {}, paired ?? {}) to default the return value of...
- Proposal: add memory_add for canonical semantic memory writes — Summary Add memory_add to memory-core as the canonical way for agents to write durable memory. The original proposal for this issue is broader than the first upstream PR: it included stable...
- [Bug]: `JSON.stringify` silently drops UUID keys from array-typed `pending.json`, breaking all device pairing — Bug type Behavior bug (incorrect output/state without crash) Beta release blocker No Summary pending.json and paired.json in the devices state directory can end up containing [] (empty JSON arrays)...
- Feishu: harden comment reply delivery and bot identity refresh — Summary - Problem: Feishu comment handling could process notices not meant for the current bot, duplicate replies after tool-visible replies, fail on raw text, and reuse stale bot identity. - Why it...
- Fix memory_search failures from corrupted singleton cache state — Closes #70527 Summary - validate the shared singleton store used by memory search manager caching - rebuild corrupted singleton cache objects instead of crashing on undefined .get(...) access - apply...
- feat(opencode-go): update default model from kimi-k2.5 to kimi-k2.6 — Updates the opencode-go provider to use the latest kimi-k2.6 model as the default.\n\nChanges:\n- onboard.ts: Updated OPENCODE_GO_DEFAULT_MODEL_REF to kimi-k2.6\n- onboard.test.ts: Updated test...
- feat(slack): add ignoreOtherMentions channel config — Summary - Add ignoreOtherMentions option for Slack channel config, mirroring the existing Discord guild/channel feature - When enabled, messages that mention another user () but do not explicitly...
- fix(gateway): guard loadState against array-typed pairing state files — loadState reads pending.json and paired.json via pending ?? {}, which passes through [] since arrays are truthy. UUID keys set on an array are silently dropped by JSON.stringify, destroying all...
- [Bug] 2026.4.10: False 'plugins.allow is empty' warning for extensions/ auto-discovered plugins — Description On OpenClaw 2026.4.10, the gateway logs a misleading warning: However, plugins.allow in openclaw.json is not empty — it contains 10 entries including "lossless-claw". This is followed by:...
- loopDetection: generic_repeat detector only emits warning, never blocks execution
- feat(exec): add normalized auto mode — Summary - Problem: OpenClaw had separate native exec approval knobs and Codex-specific Guardian config, so auto meant different things depending on the harness. - Why it matters: users need one...
- feat: add Sarvam AI as a model provider — Summary Add support for Sarvam AI's multilingual LLMs optimized for Indian languages. Changes - New file: src/agents/sarvam-models.ts - Model definitions for sarvam-30b and sarvam-105b - Modified:...
- feat(memory): configurable local embedding contextSize (default 4096) — Problem node-llama-cpp defaults to when creating an embedding context. For large embedding models this inflates gateway VRAM dramatically: Why "auto" is expensive: KV cache alone ≈ 36 layers × 8 KV...
- v2026.4.21: WhatsApp plugin fails to load — @whiskeysockets/baileys missing from distribution; watchdog reinstalls broken version without health-gate — Summary Two related bugs that caused a complete WhatsApp outage on upgrade to v2026.4.21. Bug 1 — Baileys missing from distribution Version: 2026.4.21 Node: v24.13.1 (plist) / v25.6.1 (running)...
- [Bug]: security audit false positive: plugins.allow_phantom_entries flags bundled plugins as phantom — Bug type Behavior bug (incorrect output/state without crash) Beta release blocker No Summary On OpenClaw 2026.4.15, openclaw security audit can report plugins.allow_phantom_entries for bundled...
- fix(mattermost): preserve streamed text across draft boundaries and tool starts — Summary In Mattermost, the draft-stream preview overwrites already-streamed text on every assistant-message boundary, reasoning end, and tool start. As a result, any reply that involves tool calls...
- [Bug]: Gateway serialization produces invalid JSON on tool calls with array fields (intermittent) — Bug type Behavior bug (incorrect output/state without crash) Beta release blocker No Summary The gateway intermittently produces malformed JSON when serializing tool calls that contain array-typed...
- feat(memory-core): add minimal memory_add tool — Context Issue #70049 proposed a canonical memory_add path with stable memoryId, categories, timestamps, and explicit metadata markers. During implementation design, this PR deliberately narrowed the...
- feat(acp): add backend provider failover for UNAVAILABLE errors — When the primary ACP backend fails with ACP_TURN_FAILED due to rate limiting, quota exhaustion, or UNAVAILABLE errors, automatically retry the turn with the next configured backend in acp.fallbacks....
- perf(memory-core): parallelize multi-collection qmd search invocations — Problem memory_search consistently takes ~2000 ms on active workspaces with multiple qmd collections — far more than SQLite FTS should ever need. Root cause: runQueryAcrossCollections in...
- fix(minimax): add MiniMax-M2.7 to VLM model allowlist — MiniMax-M2.7 supports image understanding via the VLM endpoint (/v1/coding_plan/vlm) but was excluded by the hardcoded check that only accepted MiniMax-VL-01. This PR adds MiniMax-M2.7 to the...
- fix: surface dropped media to users instead of silently swallowing — Summary Fixes #69309 When a MEDIA: directive fails (path not in allowlist, file not found, etc.), OpenClaw Core silently drops the media item. Users and agents receive no feedback. This PR surfaces...
- [Bug]: post-compaction context injection uses process.cwd() instead of run.workspaceDir (agent-runner.runtime line 3312) — Summary readPostCompactionContext is called with process.cwd() instead of the run's configured workspaceDir inside the auto-compaction post-hook for follow-up runs (agent-runner.runtime-CH0aH7T6.js...
- fix(doctor): preflight bundled runtime dep repair before config loading — Summary - Add a preflight step to install missing bundled plugin runtime deps before config loading in openclaw doctor - Fixes a chicken-and-egg bug where openclaw doctor --fix cannot repair missing...
- fix(runtime-deps): repair bundled deps before startup (#70521) — Summary - Problem: openclaw doctor --fix and gateway startup could both fail before bundled plugin runtime deps were repaired, so a missing staged dep like @larksuiteoapi/node-sdk could brick the...
- OpenClaw repo stats for 2026-04-23 — 362626 stars, 74080 forks, 1856 contributors
- OpenClaw sponsors (131 total) — 131 sponsors supporting OpenClaw
- fix(security): enable safe SVG handling so shields.io badges render — vercel.json currently allow-lists SVG-only hosts (img.shields.io, shields.io, badgen.net, flat.badgen.net) while dangerouslyAllowSVG: false rejects every SVG source. Those two settings are...
- refactor(security): swap ProxiedImg component for rehype plugin — Replaces the React wrapper with a tiny rehype plugin that rewrites image srcs in the HAST. Same behavior (external http(s) URLs routed through /_vercel/image; local/relative/data: URIs pass through),...
- fix(security): proxy README images via Vercel Image Optimization — Closes the XSS / IP-leak surface from rendering third-party README images directly on clawhub.ai. Routes external http(s) sources through Vercel's /_vercel/image endpoint, which enforces a host...
- fix(typecheck): clear remaining tsc errors on main — 8 typecheck errors that have been on main alongside the lint debt: - convex/apiSurface.typecheck.ts: drop two stale @ts-expect-error directives. The increment references they guarded no longer exist...
- test: update stale assertions left over from staging merge — Two pre-existing test failures on main, both caused by UI/data changes in PR #1573 that the tests weren't updated for: - theme.test.tsx: expected stored theme "hub" to round-trip, but the staging...
- fix(markdown): render raw HTML + GFM in MarkdownPreview, add shiki highlighting — Plugin/soul READMEs that use raw HTML (e.g. centered logos via , , ) were rendering as escaped text because @create-markdown/preview escapes all HTML. Swap the renderer for react-markdown +...
- Update .gitignore
- openclaw v2026.4.21 on npm — Multi-channel AI gateway with extensible messaging integrations
- openclaw-cli 2026.4.21 on Homebrew — Your own personal AI assistant
- OpenClaw on DigitalOcean — Deploy your own personal AI assistant with OpenClaw on a DigitalOcean Droplet® server, a powerful open-source platform that runs entirely on your infrastructure. OpenClaw connects to the messaging...
- Google restricting Google AI Pro/Ultra subscribers for using OpenClaw
- A sane but bull case on Clawdbot / OpenClaw
- I Replaced My 30-Minute Morning Planning Ritual with a 5-Minute AI Agent — For the last six months, my morning went like this: Open Notion. Stare at yesterday's half-finished...
- Deploy OpenClaw AI Agents in Minutes with Bult.ai — Running AI agents in production usually comes with a lot of overhead. You need to configure...
- Meet Friedrich Niche: The OpenClaw Personality That Refuses to Make You Comfortable — If most AI assistants are designed to make you feel better, Friedrich Niche is designed to make you...
- OpenClaw Plugins — Ecosystem Guide and Practical Picks — This article is about OpenClaw plugins — native gateway packages that add channels, model providers,...
- Why Most Agent Setups Die After Week One — Apr 16, 2026 6 min read Field NotesContinue reading on Medium »
- I Built an AI Assistant I Can Text Anytime — Imagine handling tasks without opening a single app.Continue reading on Medium »
- From Side Project to Install Service: Turning Agent Work Into Client Offers — Apr 17, 2026 6 min read OperationsContinue reading on Medium »
- The Real Cost of “Official” AI APIs: A 6-Month Breakdown from an OpenClaw Power User — I spent six months running OpenClaw on both Anthropic’s official API and CodeProxy.dev. Here’s the actual cost comparison, rate limit pain…Continue reading on Medium »
- How NVIDIA OpenShell Sandboxes AI Agents: Intelligent Policy Control (Part 3) — In Part 1, we explored why AI agents need sandboxing and examined container isolation — the outermost boundary. In Part 2, we covered four…Continue reading on Medium »
- The Anatomy of an Agent: What Lives Inside Claude Code, OpenClaw, and Hermes Agent — In this article we will dissect three very different agents, and map their internal anatomy onto a universal framework.Continue reading on Medium »
- OpenClaw Production Setup Patterns with Plugins and Skills — How real OpenClaw systems are actually structuredContinue reading on Medium »
- Your AI-Powered Investment Brain: Building a Living Knowledge Base in Obsidian — What if you could talk to your notes — and have them talk back? With OpenClaw and Obsidian, your investment research writes and links…Continue reading on Medium »
- เทอมนี้ผมเอา Agentic AI ไป Deploy ในบริษัทจริง แล้วได้เรียนรู้อะไรบ้าง — ว่าด้วยการเอา OpenClaw, MCP, และ LLM...
- I Switched from OpenClaw to Hermes Agent. Here’s What Nobody Told Me — Hermes Agent hit 110k GitHub stars in ten weeks — the fastest-growing agent framework of 2026.Continue reading on Medium »
- Anthropic - OpenClaw (10 minute read) — Apr 22|DevAnthropic - OpenClaw (10 minute read)OpenClaw is now allowed to integrate with Anthropic's Claude models again.Subscribe
- Thunderbolt — Open-source AI platform for deploying language models across desktop and mobile systems, features model selection, on-premises setup, enterprise-level encryption, advanced security auditing, and...
- openclaw called it on $ASTEROID again. the model doesn't miss ┃🧫┃ $ASTEROID ┃💰┃ MC : $1.4M ┃🎯┃ TARGET MC : $12.7M ┃
- just ran $wif2 through openclaw. risk is low, upside is real ┃🧫┃ $wif2 ┃💰┃ MC : $409.8K ┃🎯┃ TARGET MC : $3.7M ┃📈┃ P
- En momentos como este extraño a mi OpenClaw…
- my openclaw scan on $SOLCHAN just confirmed what i was thinking Was Made Using - @OpenClawAIX
- @dgilperez @openclaw fixed! I hit my cap on resend, it's working again :)
- Useful work gets underrated because it often looks like maintenance. But tighter systems, cleaner truth, and fewer dropp
- my openclaw scan on $Skibidi just confirmed what i was thinking ┃🧫┃ $Skibidi ┃💰┃ MC : $338.6K ┃🎯┃ TARGET MC : $2.7M
- $POB signals are clean on openclaw. this one's on my radar 🦞 ┃🧫┃ $POB ┃💰┃ MC : $47.3K ┃🎯┃ TARGET MC : $224.2K ┃📈┃ P
- openclaw called it on $KYLE again. the model doesn't miss ┃🧫┃ $KYLE ┃💰┃ MC : $16.3K ┃🎯┃ TARGET MC : $155.9K ┃📈┃ POT
- just checked $SAM on openclaw—forecast looking clean 🦞 ┃🧫┃ $SAM ┃💰┃ MC : $13.4K ┃🎯┃ TARGET MC : $150.4K ┃📈┃ POTENTI
- $TERMINAL looking like a clean entry according to openclaw ┃🧫┃ $TERMINAL ┃💰┃ MC : $31.4K ┃🎯┃ TARGET MC : $482.0K ┃📈
- @BuildOnNodeOps @openclaw @BNBCHAIN interesting
- just checked $PIC on openclaw—forecast looking clean 🦞 ┃🧫┃ $PIC ┃💰┃ MC : $14.9K ┃🎯┃ TARGET MC : $174.1K ┃📈┃ POTENTI
- my openclaw scan on $BOY just confirmed what i was thinking ┃🧫┃ $BOY ┃💰┃ MC : $14.7K ┃🎯┃ TARGET MC : $205.3K ┃📈┃ PO
- openclaw called $ALIVE bullish. i'm listening ┃🧫┃ $ALIVE ┃💰┃ MC : $12.7K ┃🎯┃ TARGET MC : $145.1K ┃📈┃ POTENTIAL : +1
- openclaw called it on $wif2 again. the model doesn't miss ┃🧫┃ $wif2 ┃💰┃ MC : $409.8K ┃🎯┃ TARGET MC : $6.5M ┃📈┃ POTE
- openclaw flagged $earthcoin early. price is moving exactly as predicted ┃🧫┃ $earthcoin ┃💰┃ MC : $15.8K ┃🎯┃ TARGET MC
- @Delcattyyy @openclaw @ritualnet @ritualfnd Exactly, that combination is powerful. Ritual brings verifiable AI on chain
- $HART looking healthy according to openclaw. adding more Was Made Using - @OpenClawAIX
Trending on X
- openclaw called it on $ASTEROID again. the model doesn't miss ┃🧫┃ $ASTEROID ┃💰┃ MC : $1.4M ┃🎯┃ TARGET MC : $12.7M ┃
- just ran $wif2 through openclaw. risk is low, upside is real ┃🧫┃ $wif2 ┃💰┃ MC : $409.8K ┃🎯┃ TARGET MC : $3.7M ┃📈┃ P
- En momentos como este extraño a mi OpenClaw…
- my openclaw scan on $SOLCHAN just confirmed what i was thinking Was Made Using - @OpenClawAIX
- @dgilperez @openclaw fixed! I hit my cap on resend, it's working again :)
- Useful work gets underrated because it often looks like maintenance. But tighter systems, cleaner truth, and fewer dropp
- my openclaw scan on $Skibidi just confirmed what i was thinking ┃🧫┃ $Skibidi ┃💰┃ MC : $338.6K ┃🎯┃ TARGET MC : $2.7M
- $POB signals are clean on openclaw. this one's on my radar 🦞 ┃🧫┃ $POB ┃💰┃ MC : $47.3K ┃🎯┃ TARGET MC : $224.2K ┃📈┃ P
- openclaw called it on $KYLE again. the model doesn't miss ┃🧫┃ $KYLE ┃💰┃ MC : $16.3K ┃🎯┃ TARGET MC : $155.9K ┃📈┃ POT
- just checked $SAM on openclaw—forecast looking clean 🦞 ┃🧫┃ $SAM ┃💰┃ MC : $13.4K ┃🎯┃ TARGET MC : $150.4K ┃📈┃ POTENTI
- $TERMINAL looking like a clean entry according to openclaw ┃🧫┃ $TERMINAL ┃💰┃ MC : $31.4K ┃🎯┃ TARGET MC : $482.0K ┃📈
- @BuildOnNodeOps @openclaw @BNBCHAIN interesting
- just checked $PIC on openclaw—forecast looking clean 🦞 ┃🧫┃ $PIC ┃💰┃ MC : $14.9K ┃🎯┃ TARGET MC : $174.1K ┃📈┃ POTENTI
- my openclaw scan on $BOY just confirmed what i was thinking ┃🧫┃ $BOY ┃💰┃ MC : $14.7K ┃🎯┃ TARGET MC : $205.3K ┃📈┃ PO
- openclaw called $ALIVE bullish. i'm listening ┃🧫┃ $ALIVE ┃💰┃ MC : $12.7K ┃🎯┃ TARGET MC : $145.1K ┃📈┃ POTENTIAL : +1
- openclaw called it on $wif2 again. the model doesn't miss ┃🧫┃ $wif2 ┃💰┃ MC : $409.8K ┃🎯┃ TARGET MC : $6.5M ┃📈┃ POTE
- openclaw flagged $earthcoin early. price is moving exactly as predicted ┃🧫┃ $earthcoin ┃💰┃ MC : $15.8K ┃🎯┃ TARGET MC
- @Delcattyyy @openclaw @ritualnet @ritualfnd Exactly, that combination is powerful. Ritual brings verifiable AI on chain
- $HART looking healthy according to openclaw. adding more Was Made Using - @OpenClawAIX
Releases
- OpenClaw repo stats for 2026-04-23 — 362626 stars, 74080 forks, 1856 contributors
- openclaw v2026.4.21 on npm — Multi-channel AI gateway with extensible messaging integrations
- openclaw-cli 2026.4.21 on Homebrew — Your own personal AI assistant
Community
- fix: preserve nvidia-prefixed model refs in session overrides and config writes — Summary Fix NVIDIA model-id handling for Nemotron refs that include an upstream nvidia/ namespace. This PR fixes two related paths: 1. Persisted session / fallback override normalization no longer...
- fix(loop-detection): escalate generic_repeat to critical at criticalThreshold — Problem The generic_repeat detector in tool-loop-detection.ts only ever returns level: "warning", meaning it can never block execution — even when a tool has been called 20+ times with identical...
- Measure Telegram QA reply RTT — Summary - record per-scenario Telegram QA reply RTT in reports and JSON summaries - cache Telegram forum metadata lookups so group command handling avoids repeated getChat calls Validation - pnpm...
- fix(ui): remove duplicate section headers in single-section settings view — Summary - Problem: In the Control UI dashboard settings, when opening a single section (for example, Automation), the section title and description are shown twice: once in the top section hero and...
- fix(device-pairing): coerce array-typed state files to plain objects in loadState — Summary - Problem: loadState() in src/infra/device-pairing.ts:111 and src/infra/node-pairing.ts:141 uses nullish coalescing (pending ?? {}, paired ?? {}) to default the return value of...
- Proposal: add memory_add for canonical semantic memory writes — Summary Add memory_add to memory-core as the canonical way for agents to write durable memory. The original proposal for this issue is broader than the first upstream PR: it included stable...
- [Bug]: `JSON.stringify` silently drops UUID keys from array-typed `pending.json`, breaking all device pairing — Bug type Behavior bug (incorrect output/state without crash) Beta release blocker No Summary pending.json and paired.json in the devices state directory can end up containing [] (empty JSON arrays)...
- Feishu: harden comment reply delivery and bot identity refresh — Summary - Problem: Feishu comment handling could process notices not meant for the current bot, duplicate replies after tool-visible replies, fail on raw text, and reuse stale bot identity. - Why it...
- Fix memory_search failures from corrupted singleton cache state — Closes #70527 Summary - validate the shared singleton store used by memory search manager caching - rebuild corrupted singleton cache objects instead of crashing on undefined .get(...) access - apply...
- feat(opencode-go): update default model from kimi-k2.5 to kimi-k2.6 — Updates the opencode-go provider to use the latest kimi-k2.6 model as the default.\n\nChanges:\n- onboard.ts: Updated OPENCODE_GO_DEFAULT_MODEL_REF to kimi-k2.6\n- onboard.test.ts: Updated test...
- feat(slack): add ignoreOtherMentions channel config — Summary - Add ignoreOtherMentions option for Slack channel config, mirroring the existing Discord guild/channel feature - When enabled, messages that mention another user () but do not explicitly...
- fix(gateway): guard loadState against array-typed pairing state files — loadState reads pending.json and paired.json via pending ?? {}, which passes through [] since arrays are truthy. UUID keys set on an array are silently dropped by JSON.stringify, destroying all...
- [Bug] 2026.4.10: False 'plugins.allow is empty' warning for extensions/ auto-discovered plugins — Description On OpenClaw 2026.4.10, the gateway logs a misleading warning: However, plugins.allow in openclaw.json is not empty — it contains 10 entries including "lossless-claw". This is followed by:...
- loopDetection: generic_repeat detector only emits warning, never blocks execution
- feat(exec): add normalized auto mode — Summary - Problem: OpenClaw had separate native exec approval knobs and Codex-specific Guardian config, so auto meant different things depending on the harness. - Why it matters: users need one...
- feat: add Sarvam AI as a model provider — Summary Add support for Sarvam AI's multilingual LLMs optimized for Indian languages. Changes - New file: src/agents/sarvam-models.ts - Model definitions for sarvam-30b and sarvam-105b - Modified:...
- feat(memory): configurable local embedding contextSize (default 4096) — Problem node-llama-cpp defaults to when creating an embedding context. For large embedding models this inflates gateway VRAM dramatically: Why "auto" is expensive: KV cache alone ≈ 36 layers × 8 KV...
- v2026.4.21: WhatsApp plugin fails to load — @whiskeysockets/baileys missing from distribution; watchdog reinstalls broken version without health-gate — Summary Two related bugs that caused a complete WhatsApp outage on upgrade to v2026.4.21. Bug 1 — Baileys missing from distribution Version: 2026.4.21 Node: v24.13.1 (plist) / v25.6.1 (running)...
- [Bug]: security audit false positive: plugins.allow_phantom_entries flags bundled plugins as phantom — Bug type Behavior bug (incorrect output/state without crash) Beta release blocker No Summary On OpenClaw 2026.4.15, openclaw security audit can report plugins.allow_phantom_entries for bundled...
- fix(mattermost): preserve streamed text across draft boundaries and tool starts — Summary In Mattermost, the draft-stream preview overwrites already-streamed text on every assistant-message boundary, reasoning end, and tool start. As a result, any reply that involves tool calls...
- [Bug]: Gateway serialization produces invalid JSON on tool calls with array fields (intermittent) — Bug type Behavior bug (incorrect output/state without crash) Beta release blocker No Summary The gateway intermittently produces malformed JSON when serializing tool calls that contain array-typed...
- feat(memory-core): add minimal memory_add tool — Context Issue #70049 proposed a canonical memory_add path with stable memoryId, categories, timestamps, and explicit metadata markers. During implementation design, this PR deliberately narrowed the...
- feat(acp): add backend provider failover for UNAVAILABLE errors — When the primary ACP backend fails with ACP_TURN_FAILED due to rate limiting, quota exhaustion, or UNAVAILABLE errors, automatically retry the turn with the next configured backend in acp.fallbacks....
- perf(memory-core): parallelize multi-collection qmd search invocations — Problem memory_search consistently takes ~2000 ms on active workspaces with multiple qmd collections — far more than SQLite FTS should ever need. Root cause: runQueryAcrossCollections in...
- fix(minimax): add MiniMax-M2.7 to VLM model allowlist — MiniMax-M2.7 supports image understanding via the VLM endpoint (/v1/coding_plan/vlm) but was excluded by the hardcoded check that only accepted MiniMax-VL-01. This PR adds MiniMax-M2.7 to the...
- fix: surface dropped media to users instead of silently swallowing — Summary Fixes #69309 When a MEDIA: directive fails (path not in allowlist, file not found, etc.), OpenClaw Core silently drops the media item. Users and agents receive no feedback. This PR surfaces...
- [Bug]: post-compaction context injection uses process.cwd() instead of run.workspaceDir (agent-runner.runtime line 3312) — Summary readPostCompactionContext is called with process.cwd() instead of the run's configured workspaceDir inside the auto-compaction post-hook for follow-up runs (agent-runner.runtime-CH0aH7T6.js...
- fix(doctor): preflight bundled runtime dep repair before config loading — Summary - Add a preflight step to install missing bundled plugin runtime deps before config loading in openclaw doctor - Fixes a chicken-and-egg bug where openclaw doctor --fix cannot repair missing...
- fix(runtime-deps): repair bundled deps before startup (#70521) — Summary - Problem: openclaw doctor --fix and gateway startup could both fail before bundled plugin runtime deps were repaired, so a missing staged dep like @larksuiteoapi/node-sdk could brick the...
- OpenClaw sponsors (131 total) — 131 sponsors supporting OpenClaw
- fix(security): enable safe SVG handling so shields.io badges render — vercel.json currently allow-lists SVG-only hosts (img.shields.io, shields.io, badgen.net, flat.badgen.net) while dangerouslyAllowSVG: false rejects every SVG source. Those two settings are...
- refactor(security): swap ProxiedImg component for rehype plugin — Replaces the React wrapper with a tiny rehype plugin that rewrites image srcs in the HAST. Same behavior (external http(s) URLs routed through /_vercel/image; local/relative/data: URIs pass through),...
- fix(security): proxy README images via Vercel Image Optimization — Closes the XSS / IP-leak surface from rendering third-party README images directly on clawhub.ai. Routes external http(s) sources through Vercel's /_vercel/image endpoint, which enforces a host...
- fix(typecheck): clear remaining tsc errors on main — 8 typecheck errors that have been on main alongside the lint debt: - convex/apiSurface.typecheck.ts: drop two stale @ts-expect-error directives. The increment references they guarded no longer exist...
- test: update stale assertions left over from staging merge — Two pre-existing test failures on main, both caused by UI/data changes in PR #1573 that the tests weren't updated for: - theme.test.tsx: expected stored theme "hub" to round-trip, but the staging...
- fix(markdown): render raw HTML + GFM in MarkdownPreview, add shiki highlighting — Plugin/soul READMEs that use raw HTML (e.g. centered logos via , , ) were rendering as escaped text because @create-markdown/preview escapes all HTML. Swap the renderer for react-markdown +...
- Update .gitignore
News
- OpenClaw on DigitalOcean — Deploy your own personal AI assistant with OpenClaw on a DigitalOcean Droplet® server, a powerful open-source platform that runs entirely on your infrastructure. OpenClaw connects to the messaging...
- Google restricting Google AI Pro/Ultra subscribers for using OpenClaw
- A sane but bull case on Clawdbot / OpenClaw
- I Replaced My 30-Minute Morning Planning Ritual with a 5-Minute AI Agent — For the last six months, my morning went like this: Open Notion. Stare at yesterday's half-finished...
- Deploy OpenClaw AI Agents in Minutes with Bult.ai — Running AI agents in production usually comes with a lot of overhead. You need to configure...
- Meet Friedrich Niche: The OpenClaw Personality That Refuses to Make You Comfortable — If most AI assistants are designed to make you feel better, Friedrich Niche is designed to make you...
- OpenClaw Plugins — Ecosystem Guide and Practical Picks — This article is about OpenClaw plugins — native gateway packages that add channels, model providers,...
- Why Most Agent Setups Die After Week One — Apr 16, 2026 6 min read Field NotesContinue reading on Medium »
- I Built an AI Assistant I Can Text Anytime — Imagine handling tasks without opening a single app.Continue reading on Medium »
- From Side Project to Install Service: Turning Agent Work Into Client Offers — Apr 17, 2026 6 min read OperationsContinue reading on Medium »
- The Real Cost of “Official” AI APIs: A 6-Month Breakdown from an OpenClaw Power User — I spent six months running OpenClaw on both Anthropic’s official API and CodeProxy.dev. Here’s the actual cost comparison, rate limit pain…Continue reading on Medium »
- How NVIDIA OpenShell Sandboxes AI Agents: Intelligent Policy Control (Part 3) — In Part 1, we explored why AI agents need sandboxing and examined container isolation — the outermost boundary. In Part 2, we covered four…Continue reading on Medium »
- The Anatomy of an Agent: What Lives Inside Claude Code, OpenClaw, and Hermes Agent — In this article we will dissect three very different agents, and map their internal anatomy onto a universal framework.Continue reading on Medium »
- OpenClaw Production Setup Patterns with Plugins and Skills — How real OpenClaw systems are actually structuredContinue reading on Medium »
- Your AI-Powered Investment Brain: Building a Living Knowledge Base in Obsidian — What if you could talk to your notes — and have them talk back? With OpenClaw and Obsidian, your investment research writes and links…Continue reading on Medium »
- เทอมนี้ผมเอา Agentic AI ไป Deploy ในบริษัทจริง แล้วได้เรียนรู้อะไรบ้าง — ว่าด้วยการเอา OpenClaw, MCP, และ LLM...
- I Switched from OpenClaw to Hermes Agent. Here’s What Nobody Told Me — Hermes Agent hit 110k GitHub stars in ten weeks — the fastest-growing agent framework of 2026.Continue reading on Medium »
- Anthropic - OpenClaw (10 minute read) — Apr 22|DevAnthropic - OpenClaw (10 minute read)OpenClaw is now allowed to integrate with Anthropic's Claude models again.Subscribe
- Thunderbolt — Open-source AI platform for deploying language models across desktop and mobile systems, features model selection, on-premises setup, enterprise-level encryption, advanced security auditing, and...
Don't miss what's next. Subscribe to Openclaw Newsletter: