Utilisez votre profil avec OpenClaw
Open source · correspondance de profil: excellent · site officiel ↗
Où il va: Per-agent workspace files: your profile becomes the agent's SOUL.md / USER.md; each AI Team member becomes its own agent entry.
Étapes
- Check the prerequisite: Node.js 22 or newer (run `node --version`).
- Install: `curl -fsSL https://openclaw.ai/install.sh | bash`, then run `openclaw onboard --install-daemon`. The onboarding wizard configures the gateway, connects your model API key (Claude via Anthropic is the common default) and creates your first workspace; the daemon flag keeps it running across reboots.
- Open your agent's workspace folder (shown at the end of onboarding). Download your Agent of Me profile as .md and save its content into USER.md (who you are), and the HOW I WRITE / communication sections into SOUL.md (how the agent behaves). That one step makes your self-hosted assistant answer and draft like you.
- Connect a channel in onboarding (WhatsApp, Telegram, Discord…) and message your agent to confirm it responds.
- If something doesn't respond, `openclaw daemon logs` shows the actual error. It's the right first debugging step.
Running a team of agents
- OpenClaw runs multiple ISOLATED agents in one gateway, each with its own workspace, state directory and session history. That maps one-to-one onto your Agent of Me team.
- For each team member, download its prompt (.md) from your AI Team page. Create an agent per member in the config (~/.openclaw/openclaw.json) under `agents.entries`, giving each its own `workspace` folder.
- Paste each team member's composed prompt into that agent's workspace persona files (AGENTS.md / SOUL.md). Your Financial Analyst, Research Analyst and Executive Assistant now exist as separate, isolated assistants with your voice built in.
- Route messages with bindings: a binding maps a channel account (a specific WhatsApp number, a Telegram bot, a Slack workspace) to one agent, so texting one number reaches My Financial Analyst while another reaches My Executive Assistant.
- Agents don't share sessions or credentials unless you explicitly allow it, clean separation, exactly like your team page.
Conseils
- Commands and menus evolve quickly in active open-source projects, docs.openclaw.ai is authoritative; the concepts here (workspace persona files, agents.entries, bindings) are the stable part.
- Model costs are yours: the gateway is free, but every message uses your API key. Start with one agent before running the whole team.
Pas encore de profil ?
Les chemins d'interface reflètent notre dernière vérification (2026-08-10), les menus changent ; le concept (instructions persistantes ou collées par chat) reste stable.