Non-interactive mode

Use interpreter exec when you want a single prompt to run start-to-finish without the terminal UI. The final result goes to stdout. Progress streams to stderr so you can pipe the result anywhere.

interpreter exec "summarize the changes in the last commit"

Common flags

Run interpreter exec --help for the full list.

Piping in context

You can feed stdin as additional context for the prompt:

git diff | interpreter exec "explain this diff in plain English"

Or pass the whole prompt on stdin:

cat task.md | interpreter exec -

Structured output

Pair --json with --output-schema to get a structured answer your script can parse:

echo '{"type":"object","properties":{"bug":{"type":"string"},"fix":{"type":"string"}},"required":["bug","fix"]}' > schema.json

interpreter exec --json --output-schema schema.json \
  "find one bug in src/parser.rs and propose a fix"

The final stdout line is a single JSON object that matches the schema.

In CI

API keys are the right default for CI. Set them as secrets, then call interpreter exec:

- name: Triage failing tests
  env:
    OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
  run: |
    interpreter exec --json \
      "look at the failing tests in pytest output and suggest a fix" \
      < pytest.log > suggestion.json

OPENAI_API_KEY=$KEY interpreter exec --full-auto \
  "bump the version in pyproject.toml and write a CHANGELOG entry"

Resuming an exec session

Stage longer pipelines by chaining runs:

interpreter exec "draft a refactor plan for src/parser.rs"
interpreter exec resume --last "now apply the plan"

resume --last continues the most recent session.

Logging

interpreter exec defaults to RUST_LOG=error and prints inline. If you want more detail:

RUST_LOG=info interpreter exec "..."