Company brain: agent for follow-up

Runs the follow-up agent cookbook in examples/cookbooks/company_brain/follow_up_agent/. It remembers your Granola calls, Linear issues and Gmail inbox in the cognee dataset follow_up_agent, then works out the next steps of your latest call and posts them to Slack, or prints them. Setup, the scripts and what each one does are in the cookbook's README.md.

Run every command from the repo root.

Steps

With none of Granola, Linear and Gmail set up, the commands below run on sample calls, issues and email (the [setup] line says so), follow up a fictional call and never post to Slack; only LLM_API_KEY is needed. Tell the user the steps come from sample data. Add --sample to use the sample even when their accounts are set up.

1. Run uv run python examples/cookbooks/company_brain/follow_up_agent/follow_up_agent.py --check.
   • Exit 0: go to step 2. [setup] SKIPPED: lines name optional sources that won't be used; mention them to the user, but they don't block the run.
   • Exit 2: each [setup] MISSING: line names one fix. Don't create credentials yourself. Tell the user exactly which line to fix, using the README's "What it needs" table, and stop.
2. Run uv run python examples/cookbooks/company_brain/follow_up_agent/follow_up_agent.py. Add --days N if the user named a time range, and --no-linear / --no-email if they don't want a source used.
3. The steps are everything after the [follow_up] Posted to Slack: or [follow_up] Steps line. Give them to the user, say which call they come from (the [follow_up] Call: line), which sources went in (the [ingest_*] lines), and whether they were posted.
4. If a script fails, follow_up_agent.py exits 1 with a line naming what went wrong. Run that script from examples/cookbooks/company_brain/follow_up_agent/scripts/ on its own to look closer.

Rules

• Gmail access is read-only.
• With Slack set up, a run posts to the user's real channel, and every run reads the user's real calls, issues and mailbox and uses LLM credits, so run it only when the user asked for it.
• Don't print the contents of credentials.json, token.json or .env.
• A sample run forgets the cookbook's dataset follow_up_agent before it ingests (the [clear] line). On the user's own data, pass --clear only when they asked to start over: it forgets everything the dataset holds.
• Don't pass --ui unless the user asked to browse the graph: it keeps running until Ctrl+C.

Clean up

uv run cognee-cli forget --dataset follow_up_agent