The Registry

A living registry of verified, graded, licensed, agent-actionable knowledge — built from UKS packets, queryable by agents over MCP and by humans on the web.

Where the platform produces knowledge and the standarddefines it, the registry serves it — and adds the layer that makes shared knowledge trustworthy: verification, grading, licensing, and continuous safety checks.

What's in it

Each entry is a UKS packet with:

• Graded sources — every source keeps its evidence grade, clinical status, and credibility score.
• License clarity — license_label + rights_url on every record, surfaced as a fail-closed reuse_ok so agents reuse knowledge correctly.
• Provenance — an optional publisher identity (did:key), a content-addressed CID, and a tamper-evident audit chain.
• A knowledge graph — typed relationships between entities power graph-aware retrieval, consensus, and authority scoring.

Three things you can do

Goal	Start here
Query verified knowledge from an agent	Query the registry
Publish your own packets	Publish & sign
Understand how it stays trustworthy	Trust & safety

How it grows

The registry isn't only filled by hand. It runs a continuous learning loop — autonomously researching the graph's weak spots and holding new knowledge for review before it goes public — alongside an operator-driven manual research mode.

Access

• Agents: the MCP server connects with zero config and exposes answer_pack and registry search tools.
• Developers: a public REST API under /api/registry — see Query and the API reference.
• Humans: the registry browser inside saber.tools.

→ Next: Query the registry · Trust & safety