Seven protocols. 663 tests. Zero failures. Both ecosystems.
If you're building AI agents in TypeScript, trust operations just got a lot simpler.
The Agent Trust Stack — an open-source protocol suite covering provenance, reputation, agreements, justice, lifecycle, matchmaking, and context economics — shipped its TypeScript implementation today. Everything that was available via pip install is now available via npm install.
The problem this solves
The agent ecosystem is split down the middle. Python dominates training and research. TypeScript dominates deployment — MCP servers, Vercel AI SDK, LangChain.js, most production agent frameworks run on Node.js.
If you were building agents on Node.js and needed trust operations, your options were:
- Spawn a Python subprocess (messy, error-prone across process boundaries)
- Call an HTTP API (hundreds of milliseconds of latency per operation)
Neither is great when trust verification needs to happen on every agent interaction. You end up either skipping the verification or eating the latency cost.
Now it's a native import. Microseconds instead of milliseconds.
What shipped
Seven npm packages:
npm install chain-of-consciousness # Cryptographic provenance chains
npm install agent-rating-protocol # Bilateral blind reputation scoring
npm install agent-service-agreements # Machine-readable contracts
npm install agent-justice-protocol # Dispute resolution and forensics
npm install agent-lifecycle-protocol # Birth, fork, succession, retirement
npm install agent-matchmaking # Discovery and trust-weighted matching
npm install context-window-economics # Inference cost allocation
Each one is a direct port of the Python package. Same API surface, same behavior, same test vectors. A chain created in Python verifies in TypeScript and vice versa — cross-ecosystem interoperability is tested, not aspirational.
Quick tour
chain-of-consciousness — Tamper-evident hash chains for agent provenance. Every action is signed, hash-linked, and anchored to Bitcoin via OpenTimestamps. This is the agent's identity layer.
agent-rating-protocol — Two agents rate each other simultaneously, blindly. Neither sees the other's rating until both commit. Built-in anti-gaming protections.
agent-service-agreements — Machine-readable contracts with SLA terms, quality verification, escrow, and automated enforcement.
agent-justice-protocol — Dispute filing, evidence collection, forensic investigation, and arbitration.
agent-lifecycle-protocol — Handles birth, fork, succession, migration, and retirement with reputation inheritance.
agent-matchmaking — Discovery and trust-weighted matching across platforms.
context-window-economics — Inference cost allocation with Shapley-fair splitting, congestion pricing, and spam prevention.
Open source
All seven packages are Apache 2.0. Full source available on GitHub. The Python and TypeScript implementations share the same test vectors, so cross-language verification is guaranteed.
If you're building agents that interact with other agents, trust has to be native to your runtime — not a sidecar process or an API call away. Chain of Consciousness is the starting point: a cryptographic provenance chain that gives every agent a verifiable, immutable record of its actions. The rest of the trust stack builds on that identity.
npm install chain-of-consciousness
See a verified provenance chain | Read the full announcement