Deep-dive technical documentation for building production-grade autonomous infrastructure.
New to self-hosted AI? Follow our curated path to go from zero to a fully integrated WhatsApp/Slack assistant in one afternoon.
Debug OpenAI-compatible relays, local-model servers, Ollama `/v1`, vLLM, LiteLLM, tools, reasoning, and streaming mismatches with our new compatibility track.
Choose a safe OpenClaw starter preset for local, coding, or remote use, install it with a rollback path, and verify the baseline before you add channels, skills, or automation.
Use the right OpenClaw CLI command for the job: map symptoms to the correct layer, run a minimal diagnosis pack, and know what evidence to capture before you change config or ask for help.
A field guide to running OpenClaw on native Windows without guesswork: separate shell, Scheduled Task, and node-host environments; fix PATH-driven failures cleanly; and know when WSL2 is the faster exit.
A practical decision guide for Windows users: choose between native Windows and WSL2 based on service model, tool compatibility, and debugging cost, then switch paths cleanly when the platform is the problem.
A practical step-by-step guide to fix OpenClaw when it suddenly stops using tools after recent updates. Learn how to check `tools.profile`, restore coding tools safely, and verify the agent can act again.
A layered checklist for the most common 'can't install' / 'command not found' / 'service won't start' failures. Covers Node version, global install permissions, PATH issues, WSL2 systemd, and Docker setup gotchas.
Harden OpenClaw against unsafe skills and prompt injection with a workflow you can actually operate: separate lab from prod, constrain blast radius, and verify that dangerous actions still require explicit trust.
A high-signal checklist for when OpenClaw stops replying (TUI shows '(no output)', channels go quiet, or logs show 401/403/429). Covers config precedence, provider auth, model allowlists, relay API-mode mismatch, and rate-limit/billing traps.
Download a ready-made TASKS.md file for your OpenClaw workspace to track multi-step work, make cron runs leave evidence, and reduce 'it promised but I can't see what happened' failures.
Deploy OpenClaw with Docker Compose, persistent host state, and a repeatable verification loop so upgrades, restarts, and channel changes do not turn into guesswork.
Set up Telegram for OpenClaw with safe DM and group controls, privacy-mode awareness, and a verification path that proves the bot sees the right messages.
Build a bounded OpenClaw operating posture before incidents happen: choose the right tool and channel limits, make every run leave evidence, and verify that your kill switch and rollback path actually work.
Choose the right remote Control UI access pattern for OpenClaw, set it up with stable origin and token behavior, and verify that pairing survives refreshes, restarts, and second-device access.
A scannable triage flow for non-hardcore operators: map what you see (unauthorized, pairing required 1008, gateway disconnected 4008, no output, only chats) to the correct fix page, with a minimal command pack to prove each layer.
A practical guide to Brave's `llm-context` mode in OpenClaw: what it returns, when to use it instead of normal web search, how to think about freshness and region, and how to avoid common key and environment mistakes.
A practical guide to OpenClaw's built-in backup commands: choose the right backup scope, verify archives before risky changes, and keep a real rollback path for upgrades, migrations, and config surgery.
Fix OpenClaw exec blocking without collapsing your trust model: separate approval layers, identify the real execution host, choose the right allowlist posture, and verify the fix cleanly.
Choose a stable remote-browser topology for OpenClaw, pin the right browser-capable node, and verify that gateway routing, relay startup, and browser takeover all land on the machine you intended.
A practical guide to OpenClaw Talk Mode's silence timeout: understand the platform defaults, decide when to shorten or lengthen the pause window, and tune for dictation, natural conversation, or slow-thinking workflows.
Choose the right OpenClaw model-serving path, validate the first backend cleanly, and know what tradeoffs you are accepting before you add tools, routing, or proxy layers.
Choose a self-hosted or proxy AI backend for OpenClaw without guessing: classify the compatibility layer, prove the runtime features you actually need, and avoid mistaking basic chat success for full agent compatibility.
A practical guide to using ClawHub as a skill discovery and distribution channel: how to evaluate maintainers, install in a low-risk way, upgrade without surprises, and keep a real rollback path when a skill fails or disappears.
A practical guide to using OpenClaw with Home Assistant without over-automating your house: where the boundary should live, what to delegate to an agent, and which risks to control first.
Design a multi-agent OpenClaw setup that stays understandable: choose what should be shared, bind every route on purpose, give each agent a real job, and verify that each ingress path lands on the worker you intended.
Decide whether Raspberry Pi should be your OpenClaw edge node or whether a NAS, Mac mini, or mini PC is the better long-term host once browser work, concurrency, and recovery matter.
Understand what Control UI pairing actually proves, why local and remote access behave differently, where users confuse pairing with auth, and how to troubleshoot trust failures without guessing.
Understand why OpenClaw recovery can keep looping after a successful OAuth renewal, and how to resync the live auth file, the provisioned copy, and safe-mode markers.
Restore stable Control UI access by separating gateway auth from device pairing, fixing unauthorized and 1008 loops, and verifying that your remote path stays stable.
Reduce OpenClaw account-ban and ToS risk by separating fragile consumer auth from production workflows, tightening blast radius, and verifying safer fallback paths before incidents happen.
Stabilize OpenClaw browser runs (Playwright) when they time out: verify the browser service is alive, increase timeouts where appropriate, reduce page complexity, and fix common headless/container dependencies.
Build a reliable OpenClaw cron and heartbeat baseline: keep the gateway truly always-on, prove runs really executed, design delivery that leaves evidence, and verify automation survives restarts and upgrades.
Set up OpenClaw email with a path you can actually keep stable: choose the right auth model, reduce Gmail re-auth loops, contain blast radius, and verify reliability before the inbox becomes business-critical.
A practical operability guide: how to capture high-signal logs, design runs that always leave evidence, and add a lightweight Kanban-style task board in your workspace so multi-step automations don’t feel invisible.
Fix OpenClaw persistence the boring way: confirm the real state and workspace paths, repair write permissions, verify artifacts survive restarts, and stop confusing runtime drift with memory loss.
A practical integration guide for using OpenClaw with OpenAI/Anthropic-compatible relays and API proxies (NewAPI, OneAPI, AnyRouter, LiteLLM, vLLM): choose the right API mode, set baseUrl correctly, avoid config precedence traps, and debug 403/404/blank-output failures fast.
A platform-aware guide for Alibaba Cloud, Volcengine, Zeabur, Unraid, AWS, and Google Cloud: why image resets force token re-entry, how to persist OpenClaw state correctly, and how to debug STARTING / startup probe failed step-by-step.
Set up openclaw.json without guesswork: confirm the active config path, choose a safe baseline, add providers and channels in the right order, and verify each change before it becomes a production problem.
A beginner-safe, execution-first checklist for your first OpenClaw week: daemon install, tokenized dashboard access, channel safety policy, backups, and fast debugging.
Update or migrate OpenClaw with a rollback plan, a full state snapshot, and a verification loop so sessions, credentials, and services survive the change.
Connect WhatsApp to OpenClaw, choose a safe inbound policy, complete QR login, and verify that only intended senders can reach the agent.
Turn on optional plugins and explicitly allowlist tools so your agent can safely call them.
Our plugin SDK is open source. Contribute your own integrations or request a dedicated build from the core team.