Tractum

released · python 3.11+ ↗︎· BUSL-1.1· shipped Oct 17, 2026

Your agents already vote. Nobody kept the ballots.

Tractum makes multi-agent decisions verifiable: every agent signs its own stance, the result ships as a k-of-n signed receipt anyone can verify, and the .curt transcripts it was computed from are bound inside the same signature.

install pip install tractum ↗︎source ↗︎homepage ↗︎

Python SDK + MCP server formk-of-n signed quorum receiptsEd25519 · ML-DSA-65 signaturesJSON · W3C VC · Nostr NIP-78 transports

Frameworks coordinate. Tractum agrees — on the record.

Tractum makes multi-agent decisions verifiable: three specialists debate, each signs its own stance, and the result ships as a k-of-n signed receipt anyone can verify — with the .curt transcripts it was computed from bound inside the same signature. Signed evidence, signed agreement, one artifact. Install with pip install tractum and get a signed receipt from any MCP-aware agent in five minutes.

Install

pip install tractumExtras: [mcp] [nostr] [pq] [langgraph] [maf] [crewai] ↗︎

Features

Every agent signs its own stanceA Tractum receipt is a k-of-n multi-sig over machine judgment, not one operator's claim about what the agents said. Each statement carries its own signature over the answer, confidence, model, role, and task — and verify_receipt_detailed counts them, cross-checks each against the response it claims to attest, and catches key substitution when you pin per-agent keys.
Decision and evidence, one signatureThe quorum receipt is the decision; the .curt transcript is the evidence. link_receipt embeds the transcript hashes inside the signed payload, so the aggregate signature covers the linkage — the evidence can't be swapped under a valid receipt. One artifact, verifiable end to end.
Any agent can call itNo framework to adopt. The MCP server exposes reach_consensus and verify_consensus to Claude Code, Cursor, Goose — one tool call gets a signed agreement, and the verifier answers statement-by-statement. Adapters for CrewAI, LangGraph, and Microsoft Agent Framework wrap the crews that already exist.

Privacy

Your prompts and answers go to exactly one place: the model provider you configure. Nothing else leaves the process. Receipts sign canonical payloads — answer, confidence, model, role, task — and verification is local. The Nostr transport is opt-in: when you publish, you publish exactly the signed receipt you chose to emit.

FAQ

Wasn't this called Claudeway?Yes — renamed in v0.5.0 (2026-10-17). The wire-format type tags are unchanged and the verifier accepts them forever: receipts published under the old name still verify.
Why not just use LangGraph or CrewAI?They coordinate; Tractum agrees. Frameworks run the agents. Tractum produces what none of them do: per-agent signed, quorum-verifiable artifacts you can hand to an auditor, a regulator, or a future teammate — with adapters that drop into those frameworks.
Why post-quantum signatures?An Ed25519 signature is 64 bytes; ML-DSA-65 (FIPS 204, NIST level 3) is ~3.3KB. Bytes are cheap; an attestation that stops being evidence the day a cryptographically relevant quantum computer lands is not. Sign big on day one — retrofit after the fact costs the whole archive.