@weaveintel/artifacts

Turns things an agent produces — a chart, a report, a code file — into stored, shareable artifacts, with redaction and sensitivity gating before anything leaves the building.

Why it exists

When an AI agent generates a spreadsheet or a document, that output needs to become a real, addressable thing: something with an id and a version, that you can store, fetch, and hand to someone else. But not everything is safe to share — a generated report might contain an API key or a customer's email. Think of it as the mailroom: it packages the parcel, stamps a version on it, and screens the contents before it goes out. This package is that mailroom — it models an artifact, stores it, and redacts or blocks content based on how sensitive it is.

When to reach for it

Reach for it whenever an agent's output should be persisted and possibly shared: generated files, images, code, structured results. Use redactText and the policy helpers before publishing anything that could carry secrets or PII. Start with the in-memory store for tests and the SQLite or filesystem backends for real storage. If you only need to remember facts across a conversation, that's @weaveintel/memory; artifacts are for the concrete outputs you keep and pass on.

How to use it

import { createArtifact, createInMemoryArtifactStore, redactText } from '@weaveintel/artifacts';

const store = createInMemoryArtifactStore();

const { text, redactions } = redactText('Deploy key: sk-live-abc123', 'secrets');

const artifact = createArtifact({
  name: 'deploy-notes.txt',
  type: 'document',
  mimeType: 'text/plain',
  data: text, // secret scrubbed before storage
});

await store.save(artifact);
console.log(`stored ${artifact.id}, redacted ${redactions.length} item(s)`);

What's in the box

• Model — createArtifact, createArtifactVersion, estimateSize, inferMimeType/detectImageMime/inferCodeMime.
• Stores — createInMemoryArtifactStore, plus SQLite and filesystem backends via the store factory.
• Redaction & sensitivity — redactText (secrets/PII levels), publishPolicyForSensitivity, and the policy gating helpers.
• References & streaming — reference helpers for pointing at artifacts, and streaming helpers for large payloads.

License

MIT.