Guide · Agent skill
How to use the Company brain Q&A agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The company-brain-company-qa skill teaches an AI agent to load a company's SQL database, support ticket exports and document folder into one cognee memory graph, then answer questions that need those sources together. It is for engineering and ops teams who want one searchable view across their data, and for developers who want to test cognee's company-memory cookbook before adapting it. After this guide you will know what the skill does, how to install it, how to trigger it with plain requests, and what it needs before it runs.
Company brain Q&Atopoteretes/cognee · Apache-2.0 Download the skill Quick start
- 1Get the skill. Download the ZIP from this page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder, or upload it under Settings → Capabilities → Skills in the Claude apps.
- 2Prepare the requirements. Install Python with uv, check out the cognee repository with the company_brain cookbook, and set your LLM API credentials in .env.
- 3Ask for the setup check. Describe your sources in a request, or ask for the sample company. The agent runs the check command from the repo root.
- 4Fix any missing setup line. If the check exits with code 2, fix the line it names yourself, such as a missing credential, then run the check again.
- 5Ingest and ask your question. The agent runs the full command with your question, and prints the answer after A: along with the ingest lines.
- 6Ask follow-up questions. Later questions use the existing company_qa memory without ingesting the sources again.
- 7Clean up when done. Run the cognee-cli forget command for the company_qa dataset to remove the stored memory.
What an agent skill is and what this one does
An agent skill is a folder containing a SKILL.md file. The file tells an AI agent when to use the skill and what steps to follow. This skill's SKILL.md points the agent at the company brain cookbook in the cognee repository, which holds the scripts that do the work.
The skill builds one company memory in the cognee dataset company_qa. Each source is extracted with one graph model, so the same person, team or project becomes a single linked node across the database, tickets and docs. The agent can then answer a question such as who owns or is handling something across those sources. The skill runs locally, reads the sources and writes only to cognee's memory.
Install the company-brain-company-qa skill
Download the ZIP for this skill from this page and unzip it. For a personal install, place the folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.
The skill's source lives at .agents/skills/company-brain-company-qa in the cognee repository, and its only bundled file is SKILL.md.
- Personal: ~/.claude/skills/
- Project: .claude/skills/ inside your project
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
Prepare the requirements
The skill does not create credentials or install the stack for you. You need Python with uv installed, a cognee repository checkout that includes the company_brain cookbook, and LLM API credentials set in a .env file. The skill runs every command from the repo root.
A database URL is optional and must have access to the tables you choose. Ticket exports (JSON or CSV) and a docs folder are also optional. If you have none of these, the skill can run on the built-in sample company.
- Python with uv installed
- cognee repository checkout with the company_brain cookbook
- LLM API credentials configured in .env (you set these yourself)
- Optional: a database URL with access to the chosen tables
- Optional: ticket export files (JSON or CSV) and a docs folder
Trigger the skill with natural requests
You do not need a special command. Ask in plain language. The skill is meant for requests to connect a company's database, tickets and docs, or for questions about who owns, decided or is handling something across them.
The agent will ask you for sources if you have not given them. If you have none, or want to see it work first, ask for the sample run.
- "Connect our Postgres tables, the Zendesk export and the policy docs folder into a company memory."
- "Who is handling the billing outage across our tickets and docs?"
- "Run the company brain on the sample company and show me how it works."
- "Who decided on the new refund policy, and who owns it now?"
Run a check, ingest and ask
The agent first runs a check with the sources you provide. If the check passes, it runs the same command without the check flag and adds your question. The answer is printed after the A: label, and the ingest lines show which sources went in.
Later questions do not need the sources again. The agent uses a separate ask script on the existing memory. Every run uses LLM credits, so the agent ingests only when you ask for it. If you want to browse the graph, the skill can open a UI, but that keeps running until you press Ctrl+C.
- Check: exit 0 means setup is fine; exit 2 means a setup line names one fix
- Ingest and ask: runs the full pipeline with your question
- Follow-up questions use the existing dataset
uv run python examples/cookbooks/company_brain/company_qa/company_qa.py --checkSafety notes: scripts and network use
This skill runs scripts and uses the network. Every ingest sends your source content to an LLM provider for extraction, so only use it on data you are ready to share with that provider. The skill does not print your .env file or a database URL's password, and it quotes source data only as far as needed to answer.
To forget the sample dataset, the skill runs a cleanup command. On your own data, it passes the clear option only when you ask to start over, because that forgets everything in the dataset.
- Extraction sends content to the LLM provider you configure
- Credentials are never created by the skill and are not printed
- --clear forgets everything in company_qa; use only when you ask to start over
- Clean up with: uv run cognee-cli forget --dataset company_qa
uv run cognee-cli forget --dataset company_qaTips for better results
Start with the sample company to confirm the setup works before you point the skill at real data. Name the tables you want with the tables option instead of loading a whole database.
Ask questions that need more than one source. A question answerable from a single file is better handled by a simple search.
- Run the sample first to check your setup
- Limit database loads to specific tables
- Ask cross-source questions such as ownership or decisions
- Ask a single-file question with ordinary search instead
What people use it for
- Finding who owns an open issue that touches both a database table and a design doc.
- Tracing which team made a decision recorded in a support ticket and described in a policy document.
- Checking whether a project mentioned in the docs has related tickets and data records.
- Trying the company memory workflow on the built-in sample company before adapting it to real sources.
Frequently asked questions
How do I install the company-brain-company-qa skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for a personal install or into a project's .claude/skills/ folder. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.
What do I need to run the company brain Q&A skill?
You need Python with uv installed, a cognee repository checkout that includes the company_brain cookbook, and LLM API credentials configured in .env. A database URL, ticket exports and a docs folder are optional depending on what you want to load.
Is the company-brain-company-qa skill free to use?
The skill is licensed under Apache-2.0. Each run uses LLM credits from the provider you configure, so the cost depends on that provider and how much data you ingest.
Does my company data leave my machine?
The skill runs locally, but ingesting data sends content to the LLM provider for extraction. Only ingest data you are ready to share with that provider.
Can I use this without my own data?
Yes. If you pass no sources, the run uses the built-in sample company, and the agent tells you the answer comes from sample data.
What should I do if the setup check fails?
Run the check command and read the lines marked [setup] MISSING. Each one names a single fix, such as a missing credential or dependency. The skill does not create credentials for you, so fix the named item and run the check again.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use this skill if you are an engineering or ops team that wants one searchable view across a database, support tickets and docs, or a developer who wants to try cognee's company-memory cookbook first. Skip it if your data is too sensitive to send to an LLM provider, if a simple search over one file will do, or if you do not have Python tooling and a configured cognee setup.
Download the skill