Open specification for portable, vendor-neutral AI grounding. The five-primitive protocol (factlet, FactMap, Factbook, FactSignal, low-FactSignal warning) defines how private team knowledge flows into LLM answers and how to measure when grounding coverage runs out.
- SPEC.md — the v0.1 specification (RFC 2119 normative language, ~14 sections)
- schema/factlet.schema.json — JSON Schema for individual factlet records
- schema/factbook.schema.json — JSON Schema for the Factbook container
- examples/payments-factbook.yaml — complete worked example
- Discussion / question → open a GitHub Discussion
- Substantive change (new field, new primitive, breaking schema) → open an RFC PR using rfcs/0000-template.md
- Spec text typo or ambiguity → open an Issue with section number and proposed clarification
- Security issue → see SECURITY.md, do not open a public issue
See CONTRIBUTING.md for the full RFC process.
v0.1 draft. v0.2 ships within 90 days incorporating community feedback. Pre-v1.0 minor versions MAY include breaking changes; after v1.0, breaking changes require a major version bump.
- factlet-ai/reference-sdk — reference implementations in Python and TypeScript
- factlet-ai/registry — community-contributed example Factbooks across domains
- factlet.ai — website + getting-started prompts
- factlet.ai/llms.txt — LLM-discovery file
MIT — see LICENSE.