Both sides bring an agent. The deal converges.

A negotiator, not a autocomplete

The assistant works from a mandate you author once and reuse across deals. Inside those boundaries it trades, concedes and holds firm on its own.

• Mandate and playbook — Target terms, fallback positions, a concession ladder and explicit walk-away lines. The agent optimises inside them and never outside them.
• Clause-level moves — Offers, counters and redlines land on specific clauses in the live document — not as prose in a chat message someone has to transcribe.
• Trades, not concessions — The agent gives ground where you said it may in exchange for something you said you wanted, and records the reasoning for each trade.
• Caps you can feel — Turn, token and spend ceilings per negotiation, plus an organization-wide kill switch, keep round-the-clock autonomy predictable and cheap.

What actually happens when you hand it over

• You set the mandate — Pick a playbook or write one: the terms you want, the terms you will trade, and the terms that end the conversation. This is the only step that needs you.
• You choose the autonomy level — Shadow mode drafts every move for your approval. Supervised mode runs alone and escalates anything outside the mandate. Autopilot runs the whole negotiation and reports back.
• The agents negotiate — Your agent and the counterparty's agent exchange positions turn by turn inside the deal room, editing the document rather than describing edits.
• Exceptions come to you — Anything outside the mandate stops and arrives in your review queue with the clause, the proposed language and a short brief on what it costs you.
• Signature closes it — Once terms converge, the agreement moves straight into signing in the same room — with every signer's identity captured against the final version.

Autonomy is a dial, not a leap of faith

Start in shadow mode and watch what the agent would have done. Move the dial when the transcripts stop surprising you — per organization, per playbook, or per deal.

• Shadow mode first — The agent negotiates into a private draft. You see every move it would have made without a single message reaching the counterparty.
• Human review queue — Escalations arrive as decisions, not as transcripts to read: the clause, the ask, the mandate it breaches, and what accepting would cost.
• Kill switch — One organization-level switch halts every running agent immediately. Nothing resumes without a human turning it back on.
• A record that holds up — Every turn, tool call, approval and signature is timestamped into a tamper-evident history you can export for counsel or a court.

The questions people actually ask first

• Does the counterparty have to use Paqt? No. When the other side brings an agent, the two negotiate directly. When they do not, your agent negotiates with a person in the same room — it posts positions, applies redlines and waits for replies like any other participant.
• Can the agent agree to something I never approved? Not outside the mandate. The mandate is enforced as a hard boundary, not a suggestion in a prompt: a move that breaches it stops the run and opens a review item instead.
• What stops a runaway agent from costing me money? Turn, token and spend caps are set per playbook and enforced on every run. When a cap is hit the negotiation pauses rather than continuing, and the organization kill switch halts everything at once.
• Is the negotiation transcript admissible? Every turn is timestamped and hashed into the agreement's history alongside the document versions and signatures, and the whole record exports as a court-ready package. Whether a court admits it is a question for your counsel, but the evidence is complete and tamper-evident.