v2026.4.15

Sync with upstream OpenClaw v2026.4.15

Changes

• Config template: document new Anthropic default — Claude Opus 4.7 now backs the opus alias, Claude CLI defaults, and bundled image understanding.
• Config template: document the bundled Google Gemini TTS provider (gemini-3.1-flash-tts-preview, 30 cross-lingual voices including Mandarin, 24kHz PCM/WAV output). API key auto-resolves from GEMINI_API_KEY / GOOGLE_API_KEY env vars or models.providers.google.apiKey.
• Config template: document dreaming.storage.mode default change from inline to separate — phase blocks (Light Sleep / REM Sleep) now land in memory/dreaming/{light,rem}/YYYY-MM-DD.md instead of polluting the per-day memory file. Opt-in to old behavior: plugins.entries.memory-core.config.dreaming.storage.mode: "inline".

Upstream highlights (v2026.4.15)

Quality release — mostly bugfixes, a few notable feature additions:

Features:

• Anthropic default selections, opus aliases, Claude CLI defaults, and bundled image understanding bumped to Claude Opus 4.7
• Google Gemini TTS added to bundled google plugin
• Model Auth status card in Control UI (OAuth health + provider rate-limit pressure)
• Memory/LanceDB cloud storage support
• GitHub Copilot embedding provider for memory search
• agents.defaults.experimental.localModelLean — drops heavy default tools for weak local models
• models.providers.*.models.*.compat.supportsPromptCacheKey for OpenAI-compatible proxies
• streamingWatchdogMs TUI streaming watchdog

Notable fixes:

• Dreaming storage default inline → separate (#66412)
• Unknown-tool loop guard now defaults on (stops Tool X not found loops until timeout, #67401)
• Skills cache invalidation on skills.* config writes (#67401)
• STT allowPrivateNetwork restored for self-hosted audio endpoints (#66692, regression from v2026.4.14)
• Gateway SIGUSR1 restart loop on Linux/systemd with plugin auto-enable (#67557)
• Ollama chat model ID 404 fix — strip ollama/ prefix (#67457)
• BlueBubbles catchup.maxFailureRetries ceiling (default 10) — no longer wedges on malformed messages
• OpenAI Codex legacy transport self-heals to canonical endpoint (#67635)
• Many security hardenings: SSRF, exec approval secret redaction, Matrix DM pairing, Feishu webhook, MCP loopback constant-time compare, symlink swap protection

Local config migration note

Upgrading users should consider these optional config updates (all default-compatible, no breaking changes):

1. If you reference anthropic/claude-opus-4-6 directly, consider switching to anthropic/claude-opus-4-7 — the opus alias will auto-resolve to 4.7.
2. If you use messages.tts and want to try Gemini TTS, add a providers.google block — API key auto-resolves from GEMINI_API_KEY.
3. Dreaming users: new default writes phase blocks to a separate directory. No action required unless you want to explicitly lock the behavior with storage.mode: "separate".

Full upstream changelog: https://github.com/openclaw/openclaw/releases/tag/v2026.4.15

---

Post-release fix (78c5e3e)

openclaw-update hardening — sudo doctor no longer pollutes /root:

• Previously, if a user ran a bare sudo openclaw doctor --fix (without --preserve-env=HOME), doctor ran with HOME=/root and created a ghost state directory at /root/.openclaw plus a stray systemd user service under /root/.config/systemd/user/. Those ghosts persisted and re-polluted every subsequent sudo run.
• scripts/commands/update.sh now:
  1. Cleans up any pre-existing /root ghost state before the sudo doctor call
  2. Extends the post-sudo chown to cover root-owned ~/.config/systemd/user/openclaw-*.service* files
  3. Captures doctor output to ~/.openclaw/.update-doctor.log (inside the VM) instead of /dev/null, so failures are post-mortem-able
  4. Prints a log hint after the update completes

No action required — just re-run openclaw-update to pick up the fix.

v2026.4.14

Sync with upstream OpenClaw v2026.4.14

Changes

• Gateway entrypoint compatibility: Verify both dist/index.js (canonical) and dist/entry.js (legacy) in installer and updater, ensuring safe upgrades across versions
• Startup optimization: Add systemd drop-in for NODE_COMPILE_CACHE and OPENCLAW_NO_RESPAWN (upstream recommended for VM/ARM hosts). Uses bridge pattern — auto-removed when upstream takes over natively
• New model: Add gpt-5.4-pro forward-compat entry to config template
• Doctor docs: Correct --fix/--repair documentation (they are aliases), add all upstream flags (--force, --yes, --generate-gateway-token)
• Fix plugin deps install: Swap doctor --fix execution order — sudo first (installs deps into root-owned global npm prefix), then chown, then non-sudo (config migration). Previously the non-sudo run left partial npm state that blocked the subsequent sudo install
• Uninstall cleanup: Remove systemd drop-in on uninstall

Upstream highlights (v2026.4.14)

Quality/bugfix release — no new config options required:

• gpt-5.4-pro forward-compat support
• Gateway entrypoint unified to dist/index.js in doctor/update internals
• Ollama streaming usage fix (prevents premature compaction)
• Memory/Ollama embedding adapter restored
• TTS voice note persistence fix
• UI security: marked.js → markdown-it (ReDoS fix)
• Multiple Browser/SSRF fixes and security hardening

Known community issues (v2026.4.14)

• Active Memory may cause slowdowns on some setups (#66708) — workaround: disable plugin temporarily
• Telegram setMyCommands not registering at startup (#66682) — cosmetic, bot still functional
• Subagent completion may send duplicate messages (#66519) — UX issue, no data loss

Full upstream changelog: https://github.com/openclaw/openclaw/releases/tag/v2026.4.14

v2026.4.12

Sync with upstream OpenClaw v2026.4.12

Quality release focused on reliability and new local-model options.

Changes

• Config template: Added LM Studio bundled provider example (local OpenAI-compatible models with onboarding, model discovery, and embedding support)
• Config template: Updated Codex provider notes for bundled provider + plugin-owned app-server harness
• Docs: Updated exec-policy command docs with v2026.4.12 enhancements (node-host rejection, rollback safety, sync conflict detection)
• Version bump: v2026.4.11 → v2026.4.12

Upstream highlights

• LM Studio bundled provider for local/self-hosted models
• Codex bundled provider with native threads, model discovery, and compaction
• Plugin loading narrowed to manifest-declared needs (faster, safer startup)
• Active Memory: QMD recall defaults to search, improved lexical fallback
• Dreaming: multiple reliability fixes (heartbeat dedup, confidence scoring, narrative cleanup, diary timezone)
• Memory Wiki: Unicode slug/filename fix for non-ASCII titles
• Security: busybox removed from safe bins, empty approver fix, shell wrapper injection detection
• Gateway: startup sequencing, WebSocket keepalive, mid-run message queueing fixes
• Telegram: approval button deadlock fix

Full upstream changelog: https://github.com/openclaw/openclaw/releases/tag/v2026.4.12

v2026.4.11

Sync with upstream OpenClaw v2026.4.11 (covers v2026.4.10 through v2026.4.11).

Upstream v2026.4.10 highlights

• Codex provider (codex/gpt-*)
• Active Memory plugin
• exec-policy CLI
• Seedance 2.0 video generation
• Per-provider allowPrivateNetwork
• heartbeat.timeoutSeconds, llm.idleTimeoutSeconds
• macOS MLX speech (experimental)
• Security hardening

Upstream v2026.4.11 highlights

• STT fix (#64766) — multipart audio transcription restored for OpenAI, Groq, Mistral
• Dreaming: ChatGPT import ingestion + Imported Insights / Memory Palace diary subtabs
• video_generate enhancements: URL-only delivery, typed providerOptions, reference audio, adaptive aspect ratio
• asyncCompletion config for direct channel delivery of async media results
• Control UI: [embed ...] rich output tag
• Feishu document comment sessions, Microsoft Teams reactions
• Ollama model discovery caching
• Codex OAuth scope fix, agents timeout alignment, MiniMax OAuth fix

Project changes

• New: openclaw-uninstall command — clean uninstall of all OpenClaw components (Gateway service, Docker containers/images, npm package, config, source code, Mac commands, PATH entries). Supports --yes for non-interactive use and --vm to also delete the VM.
• Config template: updated video_generate comments, added asyncCompletion option
• Docs: added Memory Wiki ChatGPT import section, active-memory chat commands, fixed wiki CLI paths
• Memory: STT unblocked (pending user verification after upgrade)

v2026.4.10

Sync with upstream OpenClaw v2026.4.10

Config template updates

• Codex provider: Added codex/gpt-* model path (Codex-managed auth, native threads, model discovery, compaction) — separate from openai/gpt-*
• Active Memory plugin: New optional pre-reply memory sub-agent that automatically surfaces relevant preferences, context, and past details (plugins.entries.active-memory)
• Seedance 2.0 video generation: Added fal/Seedance 2.0 to video generation model references
• Heartbeat timeout: Added heartbeat.timeoutSeconds for per-agent heartbeat turn timeout
• web_fetch SSRF policy: Added ssrfPolicy.allowRfc2544BenchmarkRange for fake-IP proxy environments (198.18.0.0/15)
• Per-provider allowPrivateNetwork: Added models.providers.*.request.allowPrivateNetwork for trusted self-hosted OpenAI-compatible endpoints

CLI command docs

• exec-policy: New openclaw exec-policy show|preset|set command for managing local exec approval policies

Notable upstream changes (not config-impacting)

• macOS Talk: experimental local MLX speech provider
• LLM idle timeout default extended to 120s
• Extensive security hardening (browser, sandbox, tools, exec, hooks)
• Microsoft Teams message actions, Matrix MSC4357 live markers
• QA test lanes for Matrix, Telegram, and multipass

Full upstream changelog: https://github.com/openclaw/openclaw/releases/tag/v2026.4.10

v2026.4.9

Sync with upstream OpenClaw v2026.4.9.

What's new (openclaw-orbstack)

• Installer UX: replace dot progress ticker with animated braille spinner, clean SIGTERM shutdown with no stray characters
• Config template: agents.defaults.llm.idleTimeoutSeconds — decouple LLM idle watchdog from general timeout; cron runs skip idle watchdog automatically
• Docs: openclaw memory rem-harness --path for historical daily-note backfill

Config fixes (v2026.4.9 validator cleanup)

• Template: removed 4 deprecated keys rejected by v2026.4.9 validator — rateLimitedProfileRotations, imageGenerationModel, postIndexSync, notifyUser
• Template: updated channels.telegram.streaming from scalar "partial" to object { mode: "partial" } format (doctor auto-migrates legacy scalar)
• Update script: openclaw doctor --fix now runs twice — first as user (config migration + systemd), then as sudo (bundled plugin runtime deps like @discordjs/opus that need root for global npm prefix), followed by chown to restore file ownership

Upstream highlights (OpenClaw v2026.4.7 → v2026.4.9)

• Memory/dreaming: grounded REM backfill lane, diary commit/reset flows, short-term promotion integration, heartbeat trigger acceptance, Control UI diary timeline view
• Inference hub: openclaw infer command family (models/media/web/embedding)
• Memory-wiki restored: sync/query/apply/digest/lint + claim-health + contradiction clustering
• Session checkpoints: openclaw sessions checkpoint/branch/restore for compaction snapshots
• Pluggable compaction: agents.defaults.compaction.provider for custom compaction plugins
• System prompt override: agents.defaults.systemPromptOverride for controlled prompt experiments
• Security: browser SSRF re-check after interaction navigations, dotenv blocks runtime-control env vars, node exec event sanitization, workspace plugin auth isolation, env sanitization (Java/Rust/Cargo/Git/K8s/cloud creds blocked), SHA-256 ClawHub plugin verification
• Providers: OpenAI defaults reasoning effort high; Ollama native /think support; Arcee AI (Trinity catalog); Gemma 4; OpenRouter slash prefix fix; Z.ai vendor codes 1311/1113
• Channels: npm packaging mirrors bundled channel sidecar deps, Matrix sync readiness, Slack bearer auth fix, ANNOUNCE_SKIP/REPLY_SKIP/NO_REPLY token leak fixes, session routing preservation, Telegram errorPolicy/errorCooldownMs
• Tools: tools.experimental.planTool opt-out, video_generate/music_generate built-in tools
• Agents: agents.defaults.params, idle timeout inheritance, /reset clears auto-fallback model overrides
• Stability: webhook ingress plugin, heartbeat prompt-section controls, Plugin SDK command-auth split subpath

Full upstream changelogs:

• v2026.4.7
• v2026.4.8
• v2026.4.9

v2026.4.8

Sync with upstream OpenClaw v2026.4.8 (covers v2026.4.7 through v2026.4.8).

Config template

• agents.defaults.compaction.provider — pluggable compaction provider (replace built-in LLM summarization with a plugin)
• agents.defaults.systemPromptOverride — replace the entire assembled system prompt with a fixed string for controlled experiments
• tools.experimental.planTool — opt out of the update_plan structured progress tool (auto-enabled for OpenAI models)

CLI docs

• openclaw infer — new command hub for provider-backed inference workflows (models, media, web, embedding)
• openclaw memory wiki — restored memory-wiki stack with sync/query/apply/digest/lint operations, claim-health linting, and contradiction clustering
• openclaw sessions checkpoint/branch/restore — session compaction checkpoints with branch and restore actions

Upstream highlights (v2026.4.7)

• Arcee AI provider (Trinity catalog)
• Gemma 4 model support (Google)
• Ollama vision capability detection
• Webhook ingress plugin for external automation via TaskFlows
• Session compaction checkpoints and branch/restore in Control UI
• Heartbeat prompt-section controls
• Memory dreaming: session transcript ingestion + per-day corpus notes
• ACPX bumped to 0.5.1
• Environment variable sanitization security hardening
• SHA-256 verification for ClawHub plugin archives

Upstream highlights (v2026.4.8)

• Bundled channel sidecar fixes (Telegram, Feishu, IRC, Matrix, Slack, etc.)
• Slack Socket Mode proxy support + token resolution fix
• DNS pinning skip for trusted proxy mode
• Plugin metadata version alignment

v2026.4.5

Sync with upstream OpenClaw v2026.4.5

Breaking

• Legacy config aliases removed: talk.voiceId/talk.apiKey, agents.*.sandbox.perSession, browser.ssrfPolicy.allowPrivateNetwork, hooks.internal.handlers, and channel/group/room allow toggles replaced by canonical paths and enabled. Load-time compatibility and openclaw doctor --fix migration are preserved upstream.
• Updated all channel config examples (allow: true → enabled: true) in Discord guilds and Slack channels.

New Features (config template)

• Video generation: videoGenerationModel config + video_generate built-in tool (xAI grok-imagine-video, Alibaba Wan, Runway)
• Music generation: musicGenerationModel config + music_generate built-in tool (Google Lyria, MiniMax, Comfy workflow)
• Memory dreaming (experimental): 3-phase background memory promotion (light/deep/REM) via memory-core plugin. Config path: plugins.entries.memory-core.config.dreaming (fields: enabled, frequency). Supports /dreaming status|on|off|help chat command, cron scheduling (default 0 3 * * *), output to DREAMS.md and MEMORY.md.
• Channel context visibility: channels.defaults.contextVisibility (all | allowlist | allowlist_quote)

New Features (commands)

• openclaw config schema — enriched JSON Schema with titles and descriptions
• openclaw plugins install --force — replace existing plugin targets
• openclaw memory promote [--apply] [--limit N] — dreaming promotion
• openclaw memory promote-explain "query" — explain promotion decisions
• openclaw memory rem-harness — REM preview tooling
• /dreaming status|on|off|help chat command
• flows and tasks added to advanced commands list

New Providers

• Qwen, Fireworks AI, StepFun, Amazon Bedrock Mantle, MiniMax TTS/Search, Ollama Web Search

Other Upstream Highlights

• Control UI multilingual (12 languages)
• ComfyUI bundled comfy workflow media plugin
• ACPX runtime embedded in bundled plugin + reply_dispatch hook
• Prompt caching improvements (deterministic MCP tool ordering, normalized fingerprints)
• Claude CLI loopback MCP bridge + stdin stream-json streaming
• ClawHub search/install in Control UI Skills panel
• Matrix + iOS native exec approval prompts
• Removed: bundled CLI text-provider backends + agents.defaults.cliBackends
• Gateway stale lock auto-detection with PID recycling support

Internal

• Updated sync-upstream skill to cover docs/commands.md and fix Docs Snapshot location reference
• Fixed dreaming config: path is plugins.entries.memory-core.config.dreaming, only enabled + frequency fields (no timezone)
• Updated troubleshooting docs for v2026.4.5 lock detection improvements

v2026.4.2

Sync with upstream OpenClaw v2026.4.2

Breaking Changes

• Firecrawl config migration: tools.web.fetch.firecrawl.* → plugins.entries.firecrawl.config.webFetch.*
• xAI x_search config migration: tools.web.x_search.* → plugins.entries.xai.config.xSearch.*
• Both can be auto-migrated via openclaw doctor --fix

New Config Options

• agents.defaults.compaction.notifyUser — opt-in compaction start notice (default off)
• before_agent_reply plugin hook — short-circuit LLM with synthetic replies after inline actions

Exec Regression Fixes

v2026.4.2 fixes multiple exec regressions since v2026.3.31:

• #59092: fix loopback "pairing required" error
• #59555: fix subagent scope-upgrade pairing
• #59112: fix exec-approvals.json normalization
• Exec defaults changed to YOLO mode (security=full, ask=off)

Full upstream changelog: https://github.com/openclaw/openclaw/releases/tag/v2026.4.2

v2026.4.1

Sync with upstream OpenClaw v2026.4.1

New config options (commented examples in template)

• agents.defaults.params — global default provider parameters
• auth.cooldowns.rateLimitedProfileRotations — cap same-provider auth-profile retries
• gateway.webchat.chatHistoryMaxChars — chat history text truncation
• channels.telegram.errorPolicy / errorCooldownMs — delivery error suppression
• channels.whatsapp.reactionLevel — agent reaction guidance
• talk.voiceWake — macOS Voice Wake for Talk Mode
• searxng added as web_search provider option

Breaking migration note

• Telegram groupMentionsOnly → groups["*"].requireMention (template already uses new format)

Upstream highlights

• /tasks command for chat-native background task board
• SearXNG web search provider plugin
• Bedrock Guardrails support
• Feishu Drive comment-event flow
• Z.AI glm-5.1 + glm-5v-turbo models
• Exec approval fixes (likely resolves #58691 regression from v2026.3.31)
• 30+ bug fixes across channels, sessions, plugins, and sandbox

Full upstream release notes: https://github.com/openclaw/openclaw/releases/tag/v2026.4.1