Build Audita from the repository root with Go 1.27 or newer. The commands below assume the executable is on PATH. Each project selects its main file from its own directory, which also determines companion resources and overlay context. All models, credentials, names, and endpoints are illustrative.

Local processing

From examples/local, edit prompts.yml to match a reachable OpenAI-compatible local service and its model. Then run:

audita config validate --config config.yml
audita process --config config.yml --output corrected.json transcript.json

Processing makes model requests. Reports are written beside the selected output unless disabled. For an offline validator preview, from the same directory:

audita promptkit show prepared audita/validators.meaning_reversal_review \
  --config config.yml --profile local \
  --var "PayloadJSON=$(cat validation-candidate.json)" \
  --var 'TranscriptDescriptionBlock='

To preview the repair instruction and its output schema without inference:

audita promptkit show prepared audita/modules.grammar.repair \
  --config config.yml --profile local \
  --var 'RepairRequestJSON={"proposal_ref":1,"correction":{"id":1,"original_text":" ","corrected_text":"","confidence":0.9},"findings":[]}'

These previews render supplied data without inference. The native command help and CLI reference describe additional inspection and validation.

Service and overlay

From examples/service, edit the synthetic endpoint and models in prompts.yml. Set AUDITA_EXAMPLE_KEY to your service credential before a real request. Inspect without a provider call:

audita config validate --config config.yml --overlay testing
audita promptkit show profile service-test --config config.yml

The example does not promise a provider's model availability, price, or quality.

Edit a prompt and run again

From examples/prompt-edit, inspect the operator replacement:

audita promptkit show prompt audita/modules.grammar.proposal --config config.yml

Edit its message in prompts.yml, repeat the inspection, and run audita process --config config.yml --output corrected.json transcript.json against a configured local service. Each invocation captures the files anew.

Shared glossary library

From examples/campaigns, first.yml and second.yml each consume the common shared.yml through the companion glossary.yml. Inspect each configuration:

audita config validate --config first.yml
audita config validate --config second.yml

With a local service configured, process either campaign:

audita process --config first.yml --output first-output.json transcript.json
audita process --config second.yml --output second-output.json transcript.json

Edit prompts.yml for a local service before processing either campaign. The first uses separate party and spell corrections; the second selects only party.