OpenClaw Assistant

OpenClaw is the general-purpose Company Brain assistant. Each OpenClaw assistant is a dedicated managed runtime provisioned by erxes, connected to the model provider you choose, and reachable from the browser, from team members you invite, and optionally from Discord.

What an OpenClaw assistant is

One identifier = one runtime. Creating the identifier reserves a plan slot and starts provisioning a dedicated server for your workspace.
Isolated by design. The runtime is reached only through erxes; the browser never talks to it directly.
Provider-connected. You connect an LLM provider with an API key or a provider subscription; the credential is forwarded to your runtime and is never stored by the plugin database.
Shared with your team. Owners see all identifiers; invited members can open the runtime and check health, while manage operations (edit, members, deploy, destroy, transfer, plan selection) stay with the owner or creator.

Create an OpenClaw assistant

Open Company Brain → OpenClaw Assistant (/agent/assistant).
Select Add OpenClaw Assistant.
Choose Setup type → Create new server (or Transfer to import an existing runtime from another organization).
Enter an Identifier name and Description, then Continue.
Choose the connection type:
- API key — paste a key from a supported provider, or
- Provider subscription — use an existing subscription plan with a device-code, setup-token, or plan-key flow.
Select the provider and complete its fields, then confirm to provision.
Wait for provisioning to finish (it can take several minutes). If the provider requires an authorization step, the page guides you through it.

The identifier is created before provisioning starts, so a failed deployment keeps the identifier (and its reserved slot) for a clean retry — your stored provider choice is reused unless you change it.

Supported providers

API-key providers: Kimi For Coding, Moonshot Kimi, OpenAI, Anthropic Claude, Google Gemini, OpenRouter, DeepSeek, xAI Grok, GLM (Z.AI).

Subscription plans: ChatGPT (device-code sign-in), Claude Pro/Max (Claude setup-token), GitHub Copilot (device code), MiniMax Coding Plan (device code), Qwen Coding Plan (plan key), Z.AI GLM Coding Plan (plan key).

Each subscription option shows an in-app guide with the exact steps for that provider. Device-code providers open a provider authorization page with a one-time code; plan-key providers use a key you paste once.

Use the assistant

Chat from the assistant page. The chat iframe is health-gated: it mounts only when the runtime answers, so restarts and key changes never show raw errors.
Change provider or model from the connection dialog whenever you need to — applying keeps working even if a check times out.
Invite members so teammates can use the assistant; view-only members can open the runtime and see health.
Connect Discord from the Discord management sheet: install the official erxes AI Assistant bot for your server, then bind a channel to this assistant. Channel bindings are assistant-scoped, and response controls let you choose which messages the assistant answers.

Manage the runtime

Health is polled automatically; the detail page shows whether the runtime is serving.
Retry provisioning after a failure keeps the stored provider and credential mode unless you explicitly change them.
Restart from the toolbar when you need a fresh runtime process.
Transfer exports server credentials (name, URL, gateway token, IDs, provider, model, credential mode) that the create sheet in another organization can import. After transfer credentials are issued, destroying the source skips server teardown because the server may now be shared.
Destroy tears the runtime down. Deleting the identifier frees its plan slot.

Quotas and billing

OpenClaw and Hermes share one assistant entitlement: Business School 3, 10, and 25 People plans grant 3, 10, and 25 total assistant slots across both kinds. The limit query reports the shared limit, usage, and remaining slots; over-plan assistants use the billing selector. Enterprise installations keep their unlimited behavior.