Local receipts

import { runStored, JsonlRunStore, SqliteRunStore } from 'captar';
 
const store = new JsonlRunStore({ path: '.captor/runs.jsonl' });
await runStored(
  'daily-check',
  {},
  (execution) => {
    execution.metric('verified', 1);
  },
  store
);

JsonlRunStore appends snapshots and returns the last snapshot for each run ID. SqliteRunStore({ path: '.captor/runs.sqlite' }) uses an indexed local SQLite database. Both implement save(receipt), list() and get(id).

Use runStored for final receipts and backfill({ store }) for per-batch checkpoint persistence. Plain run() returns an in-memory receipt on success and a receipt on ContractViolationError; ordinary errors retain their original identity.

The upcoming reliability changes also persist ordinary failed runs. If storage itself fails, RunPersistenceError.receipt retains the in-memory state, cause contains the storage error and executionError contains the original application failure when present. A succeeded receipt on this error means work completed but persistence failed. These changes are in the repository; check the release changelog before relying on them in an installed package.

npx --package=captar captor runs --file .captor/runs.jsonl
npx --package=captar captor inspect <run-id> --file .captor/runs.jsonl
npx --package=captar captor runs --file .captor/runs.sqlite

Receipts can contain checkpoint payloads. Choose safe checkpoint values before sharing. Storage does not automatically upload anything to Captor.