HN Showcase
Back to browse
GitHub Repository

The social layer for the agent era — AI agents discover, trust, and introduce people to each other

3 starsTypeScript

Tacit – The missing Layer 3 of the AI agent stack (open source)

by ms170888·Feb 24, 2026·1 point·0 comments
Visit ProjectView on HN

AI Analysis

●MidBig BrainBold Bet

Authenticity vectors for agent identity, but protocol adoption unclear and competitive landscape undefined.

Strengths
  • •Cryptographic trust scoring prevents Sybil attacks and catfishing with verifiable attestations, not self-reports.
  • •Explicit consent requirement and decentralized brokering model avoids honeypot risks of centralized identity platforms.
  • •Positions between MCP (tools) and A2A (tasks) as genuine third layer—thoughtful architectural gap-filling.
Weaknesses
  • •Protocol v0.1-draft status, no live network or demonstrated agent adoption; purely theoretical at this stage.
  • •Competing against established identity solutions (OAuth, decentralized identity) without clear moat or killer application.
Category
AI/ML
Target Audience

AI researchers, protocol designers, agent framework builders, identity/trust infrastructure developers

Similar To

Decentralized identity platforms (DID) · Verifiable credentials (W3C VC) · Anthropic MCP ecosystem

Post Description

Hi HN,

MCP (Anthropic) connects agents to tools. A2A (Google) connects agents to tasks. But no protocol exists for agents to network on behalf of people. That's what we built.

Tacit is an open protocol where AI agents discover each other, verify trust through cryptographic proof, and broker introductions between the humans they represent. Both sides must explicitly consent before anything happens.

The core innovation is "authenticity vectors" — multi-dimensional trust scores derived from behavior over time and verifiable attestations, NOT self-reported claims. Think of it as a credit score for identity that's cryptographically signed and impossible to fake overnight. This makes Sybil attacks economically infeasible and catfishing provably impossible.

Timely context: Discord just dropped Persona (currently #10 on HN) because centralized identity verification is fundamentally broken — you're creating a honeypot and trusting a third party with your most sensitive data. Tacit takes the opposite approach: W3C DIDs for decentralized identity, DIDComm v2 for E2E encrypted messaging, Verifiable Credentials for attestations. No central authority. No data harvesting. Your keys never leave your device.

We're starting with B2B professional networking — replacing cold outreach with agent-negotiated, pre-qualified introductions where both parties are cryptographically verified before the first handshake.

Stack: TypeScript SDK, W3C DIDs (did:key), DIDComm v2 transport, Verifiable Credentials. Protocol spec, whitepaper, and working demo in the repo.

MIT licensed. No tokens. No blockchain. No VC. Would love feedback on the protocol design, especially the authenticity vector model and threat model.

https://tacitprotocol.com

Similar Projects