Direct your agents
Only the decisions that need you reach you.
Your agents settle the small stuff. What's left waits for you in one place, with the options and the one your agent would pick.
Without it
I spend the day answering agent questions.
A new session asks what I answered last week.
When I'm away, an agent guesses on something that mattered.
Where your answer goes
Your answer lands on the spec every agent reads.
The spec your agents open before they start
How an agent asks you.
Only about scope, customers, revenue, security, data that can't come back, a principle, or a real choice between options.
- The topic, so you know what it's about.
- The ask, one question in plain words.
- The options, with the one it would pick.
You stay in control
You make the call.
Answer in the app or from your terminal. Save an answer as a principle and it isn't asked again. There's one open ask per priority, and the agent keeps working on anything that doesn't depend on your answer.
What's inside
- Who decides
- A person makes the call. Annsa ranks and suggests.
- Agents
- Claude Code, Cursor, Codex and Grok, over MCP
- MCP tools
- Four: annsa.priorities, annsa.spec, annsa.act and annsa.ask
- Sign-in
- OAuth in the browser on the first run. There is no token file.
- Rate limits
- 500 an hour in the app, 1,000 an hour over MCP
Questions people ask.
Q1
When does an agent ask me?
When a choice changes scope, touches customers, revenue, security or data that can't come back, goes against a principle, or is a real choice between options. It settles the rest.
Q2
Can I answer from my terminal?
Yes. Answer in the app or from your terminal, and the agent reads your answer on the spec.
Q3
How many questions can an agent ask at once?
One open ask per priority. The agent keeps working on anything that doesn't depend on your answer.
Q4
What does saving an answer as a principle do?
It means the same question isn't asked again, by that agent or the next one.