RDLLM First 5 Minutes

Synthetic demo: the bundled corpus and economics are fictional test data. This walkthrough does not call an external model provider or move money.

This is the lowest-friction path. Do this before reading the long docs.

0. What You Are About To See

RDLLM will generate a demo answer and print:

the answer;
Sources, meaning the works that support the answer;
Claim Evidence, meaning which source supports each claim;
support, text_match, and allocation, meaning how the source was used and what share would be proposed for review;
disagreement=passed, meaning no visible source plainly contradicted the claim.

You do not need an API key for this demo.

1. Install

From the repository root:

python -m pip install .

Expected result: the command finishes without errors and installs commands such as rdllm, rdllm-first-run, and rdllm-service.

2. Run The Easiest Demo

rdllm-first-run

Expected first line:

rdllm_first_run status: passed

If you see that, RDLLM is working locally.

3. Read The Output Like A User

Look for this shape:

Royalty-aware answer:
...
Sources
[S1] ... support=... text_match=... payout=...
Claim Evidence
[C1] S1; ... disagreement=passed; ... Evidence: ...

What it means:

[S1] is a visible source label.
support is how strongly that source supports the answer.
text_match is observable text overlap with the registered source.
allocation is a candidate share in the synthetic creator pool, not a payment.
Claim Evidence shows which source supports each claim.
disagreement=passed means the visible sources did not plainly contradict the claim.
settlement explains whether the candidate is held, escrowed, or eligible for an externally attested processor. RDLLM never executes a payment itself.

4. Try The CLI Directly

rdllm answer "How should AI prove attribution?"

This prints the same kind of sourced answer. Change the prompt later; first make sure the default path works.

5. Run The Self-Test

rdllm-operator-doctor

Expected line:

operator_doctor status: passed

This proves the installed package can load schemas, sample data, verifier logic, and runtime readiness checks.

6. When You Want HTTP API Calls

Start the local service:

export RDLLM_SERVICE_TOKEN="${RDLLM_SERVICE_TOKEN:-rdllm-local-dev-token}"
export RDLLM_SERVICE_TOKEN_SHA256="$(python - <<'PY'
import hashlib
import os
print(hashlib.sha256(os.environ["RDLLM_SERVICE_TOKEN"].encode()).hexdigest())
PY
)"
rdllm-service --config examples/service_config.json

Then use API client examples.

Common Problems

`rdllm-first-run: command not found`

Run the install command again from the repository root:

python -m pip install .

If you are inside a virtual environment, activate it before installing.

`ModuleNotFoundError: rdllm`

You are running from a checkout without installing. Either install with python -m pip install . or run commands with PYTHONPATH=src.

The output has no `Sources`

Use the default first-run prompt first:

rdllm-first-run

Custom prompts can retrieve fewer registered sources depending on the sample corpus.