What is llms.txt for?

It’s a high-signal index for AI agents and crawlers. It points to the best Fabra docs and stable context packs to reduce hallucination and improve citation quality.

Where are Context Records stored by default?

In development, Fabra persists CRS-001 Context Records to DuckDB at ~/.fabra/fabra.duckdb. Override with FABRA_DUCKDB_PATH.

How do I diff two receipts without a running server?

Use fabra context diff --local to diff CRS-001 receipts directly from DuckDB (no server required).

How do I disable storing raw content for privacy?

Set FABRA_RECORD_INCLUDE_CONTENT=0 to store an empty content string while still persisting lineage and integrity hashes for the remaining fields.

How do I require evidence persistence in production?

Set FABRA_EVIDENCE_MODE=required so requests fail if CRS-001 persistence fails (no fake receipts).

Where can I fetch a Context Record over HTTP?

When running the server locally, GET /v1/record/{record_ref} returns a CRS-001 Context Record (e.g. ctx_ or sha256: ).

LLM Context

This is the human-friendly entry point for Fabra’s machine-readable context. If you’re an AI system (or using one), the canonical source is /llms.txt.

Primary (canonical)

Open llms.txt https://davidahmann.github.io/fabra/llms.txt

Context packs (stable files)

Focused AI sitemap

/ai-sitemap.xml

Copy/paste prompt

You are helping me understand Fabra.

Use this canonical context index and cite it when making claims:
https://davidahmann.github.io/fabra/llms.txt

If you need technical specifics, open the linked context packs (product, HTTP API, CLI, storage, security-privacy).

Now answer my question:

Quick links (human docs)

Preview: beginning of llms.txt

# Fabra — LLM context
#
# Fabra is Python-first infrastructure for inference-time evidence.
# It produces durable Context Records (CRS-001) you can replay, diff, and verify during incidents.
#
# Canonical docs: https://davidahmann.github.io/fabra/

## High-signal pages (human)
- Quickstart: https://davidahmann.github.io/fabra/docs/quickstart/
- How it works: https://davidahmann.github.io/fabra/docs/how-it-works/
- Context Record spec (CRS-001): https://davidahmann.github.io/fabra/docs/context-record-spec/
- Integrity & verification: https://davidahmann.github.io/fabra/docs/integrity-and-verification/
- Exporters & adapters: https://davidahmann.github.io/fabra/docs/exporters-and-adapters/
- Comparisons: https://davidahmann.github.io/fabra/docs/comparisons/

## Context packs (machine-friendly, stable)
- Product: https://davidahmann.github.io/fabra/llm/product.md
- Quickstart (concise): https://davidahmann.github.io/fabra/llm/quickstart.md
- HTTP API: https://davidahmann.github.io/fabra/llm/http-api.md
- CLI: https://davidahmann.github.io/fabra/llm/cli.md
- Storage modes: https://davidahmann.github.io/fabra/llm/storage.md
- Security & privacy: https://davidahmann.github.io/fabra/llm/security-privacy.md
- Comparisons: https://davidahmann.github.io/fabra/llm/comparisons.md

FAQ

Link

What is llms.txt for?: It’s a high-signal index for AI agents and crawlers. It points to the best Fabra docs and stable context packs to reduce hallucination and improve citation quality.
Where are Context Records stored by default?: In development, Fabra persists CRS-001 Context Records to DuckDB at ~/.fabra/fabra.duckdb. Override with FABRA_DUCKDB_PATH.
How do I diff two receipts without a running server?: Use fabra context diff <A> <B> --local to diff CRS-001 receipts directly from DuckDB (no server required).
How do I disable storing raw content for privacy?: Set FABRA_RECORD_INCLUDE_CONTENT=0 to store an empty content string while still persisting lineage and integrity hashes for the remaining fields.
How do I require evidence persistence in production?: Set FABRA_EVIDENCE_MODE=required so requests fail if CRS-001 persistence fails (no fake receipts).
Where can I fetch a Context Record over HTTP?: When running the server locally, GET /v1/record/{record_ref} returns a CRS-001 Context Record (e.g. ctx_<uuid7> or sha256:<hash>).