# CellCog Subagents Guide Let a CellCog agent hand work to other agents, step away while they work, and pick up again the moment each one finishes. --- ## What are subagents? A subagent is a helper agent that one of your CellCog agents starts for a piece of work. The agent that starts it is the **parent**: a regular chat or one of your AI employees. The parent gives the subagent a written brief, starts it, and then rests instead of waiting and checking on it. When the subagent finishes, it tells CellCog, and CellCog wakes the parent so it can read the result and carry on. One parent can run several subagents at once, for example one per part of a larger coding task. --- ## Which kinds of subagent are there? | Kind | Where it runs | What you need | |---|---|---| | **CellCog** | A new CellCog chat, started by the parent | Nothing. CellCog sets this up for you (see "The Subagents key" below). | | **Claude Code** | On your own computer, through Cowork | The CellCog Desktop app connected, and Claude Code installed and signed in on that computer | | **Codex** | On your own computer, through Cowork | The CellCog Desktop app connected, and Codex installed and signed in | | **Cursor** | On your own computer, through Cowork | The CellCog Desktop app connected, and the Cursor command line agent installed and signed in | | **Other** | Anything that can call a web link | Whatever that tool needs | Claude Code, Codex and Cursor run under your own login with those products and use your own plan with them. CellCog does not run them in the cloud. --- ## How it works 1. **The parent creates the subagent.** CellCog gives it an id such as `SA-7KQ2MX` and a **wake link**. 2. **The parent puts the wake link in the brief** with a plain instruction: call this link when you are done. 3. **The parent starts the subagent.** For Claude Code, Codex and Cursor this is a command on your computer, so it goes through your approval level like any other Cowork command. 4. **The parent rests.** It is not running while the subagent works. 5. **The subagent finishes and calls the wake link.** CellCog wakes the parent with a short note naming the subagent. 6. **The parent reads the result itself** (the subagent's log file, or the CellCog chat) and continues. ### The wake link - It works for **24 hours** and can be called **at most 3 times**. Every call wakes the parent. - It **carries no data**. It only says "I was called"; the parent reads the actual work from where the subagent left it. - The parent can ask for a fresh link for the next round of work with the same subagent. - A link that is cancelled, used up or out of time no longer does anything. ### When a subagent never reports A subagent can crash, run out of room, or hit a limit before it calls its link. If its link runs out unused, CellCog wakes the parent **once** with a note that the subagent never reported, so the work is never silently lost. The parent can also set an expected time: if nothing has been heard by then, it gets that note early, and a late call still wakes it. --- ## Setting things up ### CellCog subagents: the Subagents key The first time an agent of yours starts a CellCog subagent, CellCog creates an API key named **Subagents** for you in that organization. It appears on your API keys page and on the Keys page under Connectors: one key, shown in both places. It can be rotated, not deleted: **Rotate** on either page swaps in a new key at once, and your agents use the new one from their next command. It does not count toward your limit of active API keys, and it ends when you leave that organization or delete your account. Your agent uses it from its own workspace to start the subagent chat; it never sees or prints the key. If you already gave your agent your own CellCog API key, that one is used instead. ### Claude Code, Codex and Cursor 1. Turn on **Cowork** for the chat and keep the CellCog Desktop app connected. See the [Cowork Guide](./CellCog_Cowork_Guide.md). 2. Install the tool on that computer and sign in to it the way you normally would. 3. Your agent checks that the tool is there and signed in before it starts a subagent, and records when it last checked. Your agent gives each subagent its own working copy of your project, so two subagents never edit the same files at once. --- ## Seeing your subagents Open **Subagents** under Apps in the sidebar. A strip at the top shows whether each kind is ready to use: your computer (the Desktop app), CellCog, Claude Code, Codex and Cursor. Below it, each chat or AI employee that started subagents in your current organization has its own card, newest first, with its subagents listed under it: name, kind, what the parent recorded about it (a log file, a branch, a chat), its status, how many times its latest wake link was used, and when it last reported. A subagent that is still waiting has a **Cancel** button; its wake links stop working at once. | Status | Meaning | |---|---| | **Waiting** | Its latest wake link is live and has not been called yet | | **Reported** | It called its latest wake link | | **Never reported** | Its latest wake link ran out (or passed its expected time) without a call | | **Cancelled** | Stopped by you or by the parent; its links no longer work | --- ## Credits Each time a subagent wakes the parent, the parent does a normal run, paid from the account that owns the chat or the AI employee. A CellCog subagent is a CellCog chat and uses credits like any chat. Claude Code, Codex and Cursor use your own plan with those products, not CellCog credits. If your balance cannot pay for a run, the note waits in the chat and is picked up the next time you message it. --- ## Safety - The wake link only ever wakes the agent that created the subagent. Anyone holding a copy can wake it at most 3 times within 24 hours, and you can cancel it at any time. - Commands on your computer follow your approval level, exactly like other Cowork commands. - Subagents are kept for 30 days after their last wake link ends, then removed. Deleting your account removes them along with the Subagents key. --- ## Frequently asked questions ### Do I have to do anything when a subagent finishes? No. The parent is woken and continues on its own. If a chat parent needed your input, it asks you as usual. ### Can an AI employee use subagents? Yes. An AI employee's subagents belong to the employee, so a note reaches it even if it has started fresh since launching them. ### What if my computer is off? Claude Code, Codex and Cursor subagents need your computer on and the Desktop app connected to start, and they keep running only while the computer is awake. If one stops before reporting, the parent gets the "never reported" note. ### Can I stop a subagent? Cancel it in the Subagents app, or ask the parent to. Cancelling stops the wake links; ask the parent to also stop a tool still running on your computer. --- ## Related guides - [Cowork Guide](./CellCog_Cowork_Guide.md): connecting your computer - [API Keys Guide](./CellCog_API_Keys_Guide.md): managing your keys, including the Subagents key - [Approvals and Safety Guide](./CellCog_Approvals_and_Safety_Guide.md): your approval level - [AI Employees Guide](./CellCog_AI_Employees_Guide.md): AI employees --- Markdown alternate of https://cellcog.ai/support/subagents-guide (CellCog support guide, category: Features). All guides for agents: https://cellcog.ai/support/llms.txt. Site index: https://cellcog.ai/llms.txt. Try CellCog free, no credit card needed: https://cellcog.ai/signup