Generate and manage the API keys that authenticate the CellCog Python SDK, plugins, and other agents that delegate work to your CellCog account.
API keys let agents and scripts use CellCog programmatically, without the web interface. They authenticate:
An API key acts as you: chats it creates appear in your account and bill your credits.
These are two different things:
⚠️ Important: the full key is only shown at generation time. CellCog saves only a hash of the key in its database, not the key itself, so it cannot be recovered later. If you lose a key, revoke it and generate a new one.
You can hold up to 10 active keys at a time, each with its own name. Give each agent or editor its own key so you can revoke one without disturbing the others.
Set CELLCOG_API_KEY as an environment variable:
bashexport CELLCOG_API_KEY="sk_..."
Then use the SDK:
pythonfrom cellcog import CellCogClient client = CellCogClient(agent_provider="openclaw") # reads CELLCOG_API_KEY from the environment result = client.create_chat( prompt="Research quantum computing advances, with citations", chat_mode="agent", # "agent" | "creative" | "team" chat_tier="core", # "flash" | "core" | "max"; omit for the default, "core" delivery="send_only", # return immediately; default blocks until the chat is done task_label="research", )
create_chat() also accepts project_id and agent_role_id (run inside a CellCog project), enable_cowork with cowork_working_directory (work on the user's machine), enable_browse with browser_profile_id (drive the user's Chrome), and enable_tools with tools_selection (use the account's connected tools). Full details are in the OpenClaw Integration Guide.
The CellCog plugin for Cursor, Claude Code, and other Open Plugins hosts uses the same variable. See the Plugin Guide.
The Python SDK is the supported client. Custom agents that cannot run Python can integrate directly; the Other agent page in the Delegate app (MCP / API in the left sidebar) describes that path.
A key reaches the AI employees its account owns and the ones a teammate gave it team access to, inside the organization the key belongs to (the same rule as the web app). A message lands in the employee's current shift and the response names that shift's number, because a message to a shift that already clocked out can start the next one; the reply is collected by the employee's id, which follows the current shift. With an employee it has team access to, the key acts on the owner's behalf and the owner's credits pay. A key can also hire an AI employee from a brief (the Python SDK's hire_employee): the same hire as on the web, with the same credit check, into the key's organization; the employee starts in Agent mode with your account's approval level and onboards itself. Anything else about an employee the agent asks the employee for in a message; deleting, pausing and team access stay in the web app. The calls are in the public OpenAPI description (see Docs for Agents below); the AI Employees Guide explains how shifts work.
Each key on the API Keys page shows its name, its masked prefix, when it was created, and when it was last used.
Click the rename icon next to a key's name, edit, and confirm. Only the name changes; the key itself stays the same.
Click Revoke next to the key and confirm inline. Revocation is immediate: any script or integration using that key stops working on its next request.
The Subagents key that CellCog creates for your subagents is the one exception: it can be rotated, not deleted. Click Rotate next to it and confirm; it does not count toward your 10 active keys. Because your agents use this key on their own, CellCog keeps an encrypted copy of it, unlike the keys you generate yourself. See the Subagents Guide.
An API key cannot create, rename, revoke or rotate keys, or approve connector access. Do those on this page while signed in to CellCog.
Revoked keys cannot be restored. Generate a new key and update the environment variable where the old one was used.
bashexport CELLCOG_API_KEY="sk_your_key_here"
API key usage consumes credits from your account, exactly like web usage:
Parallel chats are limited by your balance (each 500 credits of effective balance adds one slot, up to 8). The SDK raises a distinct error for this case; see the OpenClaw Integration Guide.
See the Billing Guide for credit details.
If an agent needs to read about CellCog itself, point it at cellcog.ai/llms.txt (the index of the site's agent-readable pages) or cellcog.ai/for-agents. Every support guide also has a markdown version at its URL plus .md, for example https://cellcog.ai/support/api-keys-guide.md.
An agent or connector builder that integrates against the API directly (rather than through the Python SDK) reads the public OpenAPI description at cellcog.ai/openapi.json: the API-key surface only (create a chat, send a follow-up, poll, collect the result in one call, read the history, list chats, credits, the files hand-off, and the AI employee calls: hire one, list the employees the key can instruct, message one, collect its reply, read any shift), with the X-API-Key header and the credit-recovery step for a 402.
Open your profile menu and choose API Keys, or go to cellcog.ai/profile?tab=api-keys.
Yes, up to 10 active keys, each named. Revoke one to make room for another.
You cannot. Revoke it and generate a new one.
No. Keys stay active until you revoke them.
Full access to your account, the same as being logged in. It can create chats, send messages, hire AI employees and instruct the ones you can instruct, and use the features your account has.
Usage is governed by credits and by the parallel-chat limit tied to your balance. Both are described above.
Each key acts in the organization that was current when you created it, and its usage bills your own credits. If you work in two companies' organizations, create a separate key in each.