RDLLM GitHub Start Here

RDLLM is an open-source source-attribution and creator-royalty layer for AI answers. It helps an AI product show which sources support an answer, how much each visible source contributed to the output, whether the answer is safe to display as grounded, and whether value should be paid to creators or held in escrow.

Use this page when you are new to the repository and want the shortest working path from clone to a verified source footer.

If You Only Do One Thing

python -m pip install .
rdllm-first-run

Expected first line:

rdllm_first_run status: passed

For a slower copy/paste walkthrough, use First 5 Minutes.

Choose Your Explanation Depth

Start with the public explainer if you want RDLLM in four layers: ELI5, Simple, Non-Technical, and Technical. It is available in 15 languages.

Use the project attribution map if you want to audit what RDLLM was built from: papers, standards, source code, runtime examples, schemas, proof artifacts, and verification commands.

Use the white paper if you want the full state-of-the-art concept paper: attribution problem, current research evidence, runtime architecture, allocation mechanism, threat model, and public-readiness requirements.

Who This Is For

Start In 3 Commands

python -m pip install .
rdllm-first-run
rdllm-operator-doctor

Then try the live CLI example or bootstrap an operator directory:

PYTHONPATH=src python3 -m rdllm.cli answer "How should AI prove attribution?"
rdllm-operator-bootstrap --output-dir .rdllm/operator --operator-template company --operator-name "Local RDLLM" --security-contact security@example.com --include-sample-corpus --include-reference-artifacts

The answer prints inline source labels such as [S1], visible source rows, and Claim Evidence rows with support scores, evidence span hashes, source disagreement status, and payout allocation.

Live Use Cases

Multilingual Quickstarts

These concise quickstarts cover 15 public documentation entry points for the repository. The first set follows 2026 total-speaker rankings; the expanded set adds widely used developer and regional languages.

Each language also has a matching explainer: English, Simplified Chinese, Hindi, Spanish, and Arabic, French, Bengali, Portuguese, Indonesian, Urdu, Russian, German, Japanese, Korean, and Vietnamese.

Implementation Languages

The API examples prioritize common implementation languages for AI products: JavaScript, Python, TypeScript, Java, and C#. Stack Overflow's 2025 Developer Survey reports JavaScript, Python, TypeScript, Java, and C# among the most-used general-purpose programming languages for professional developers after removing markup-only, query-only, and shell-only entries from the broader programming/scripting/markup table.

What The User Sees

A grounded answer is not just a paragraph with a citation. The display surface includes:

Short runtime excerpt:

Claim Evidence
[C1] S1; claim_hash=399e7ff82b80; support=0.900; span=399e7ff82b80; chars=0-64. disagreement=passed; agreements=S1; conflicts=none; disagreement_profile=rdllm-visible-source-disagreement/v1. Evidence: Every royalty bearing AI answer should have a provenance record.

What The Verifier Proves

The verifier checks that:

SEO And GEO Notes

For search engines, this repository keeps the first page crawlable and descriptive: direct headings, local links, a static docs/index.html, release metadata, license files, and copyable commands. For generative engines, the docs use citation-friendly source rows, short summaries, and explicit field definitions so answer systems can quote the project accurately instead of guessing from code names.

The GEO-specific rule for this project is the same as the product rule: do not optimize for being cited by overstating what RDLLM proves. The public docs say RDLLM reports observable source support and settlement allocation; hidden model-internal reliance is claimed only when provider telemetry is separately verified.

Primary references:

Next Docs