Everything the agent runs on its own
Three surfaces, one agent. The assistant executes an agreement from intake to signature, bot negotiation handles the clause-by-clause exchange, and the Model Context Protocol surface lets your own systems drive both. Start with whichever one describes the work you want to stop doing by hand.
Every agentic surface
One agent, described from each angle — the end-to-end run, the clause exchange inside it, and the protocol surface your own systems drive it through.
- AI assistant — An agent that takes the agreement from intake to signature
- Bot negotiation — Agent-to-agent bargaining inside a mandate you set
- MCP — Point your own agent at Paqt without writing API calls
What makes an agent safe to hand work to
The same four controls sit underneath all three surfaces. They are the reason an agent can be given a task rather than a suggestion.
- A mandate — The approved positions, the walk-away points and the clauses that cannot move. The agent negotiates inside it and stops at the edge.
- Clause gates — Named clauses that always come back to a person, whatever the autonomy level says. The gate is configuration, not a prompt.
- Attribution — Every action is recorded against the agent, the mandate in force at the time and the document version it applied to.
- An autonomy dial — From draft-only through propose-and-wait to execute-within-mandate, set per agreement type rather than per account.
What the agent does without being asked twice
A task handed over once should not come back for instructions at every step. This is the work that happens between a request and a signature.
- Before the exchange: Read the request and pick the template it belongs to
- Before the exchange: Draft from the approved position, not from the last deal
- Before the exchange: Flag the clauses that are already outside the mandate
- During the exchange: Answer a counter-position clause by clause
- During the exchange: Concede only inside the mandate, and say why
- During the exchange: Escalate a gated clause with the argument already written
- After agreement: Assemble the final document from the accepted clauses
- After agreement: Route it for signature in the right order
- After agreement: Record the agreement history and the deviations it contains
Questions about the agents
- Is this one agent or three? One agent, described from three angles. The assistant page covers the end-to-end run, the bot negotiation page covers the clause exchange inside it, and the MCP page covers how an external system starts and reads a run.
- Does the agent sign on our behalf? Signature routing is an agent action; the signature itself is a person's. The agent assembles the final document and sends it to the right signatories in the right order.
- What happens when the counterparty asks for something outside the mandate? The agent stops and escalates with the clause, the request, the approved position it conflicts with and a drafted response. It does not concede and it does not improvise a new position.
- Can we watch a run instead of reading about it? The assistant page carries a nine-stage trace of a real run, stage by stage, including the point where it hands a gated clause back to a person.