Choose and download the archive
- 1Open Export
Go to My Account → Privacy & data → Manage company knowledge. Select the company, then Export.
- 2Select company-owned projects
Choose the projects whose plans and saved memory you want to include. Use the selection control to include all projects. Company process context is included even if you select only some projects.
- 3Choose history and observations
Enable Include version history for older versions; otherwise the archive contains the latest source versions. Enable Include mapped external observations to add stored, dated evidence from permitted connections.
- 4Download company knowledge
The browser downloads a ZIP named for the company. Export does not call an AI provider, automatically upload the archive or consume an AI reasoning request.
Understand the archive files
| File | Purpose |
|---|---|
| README.md, schema.json | Format, limitations and guidance for interpreting the records. |
| objects.jsonl, work.jsonl | Stable work identities and selected operational context. |
| memory.jsonl | Source-linked records and the selected versions. |
| process-events.jsonl | Included company process context and events. |
| provenance.prov.json, provenance.prov.jsonld | PROV representations of recorded entities, activities, version relationships and memory use. |
| connections.json, external-observations.jsonl, permissions.json | Source metadata, optional stored observations and scope information. |
| manifest.json | Archive metadata and SHA-256 checksums for the included files. |
Know what is excluded
User-authored prose can still contain sensitive information. Review the readable archive even though credential and structured-field minimization is applied.
- Connection credentials, authorization secrets and structured contact fields.
- Private learner records and another company’s knowledge, including supplier memory merely shared with a client.
- Live provider data that has not been stored as a mapped observation.
- Missing historical provenance that was never recorded.
Prepare it for your own knowledge or LLM system
- 1Validate the package
Read README.md and schema.json. Verify file checksums from manifest.json before importing the data.
- 2Preserve scope and evidence
Keep company, project and object boundaries, version identities and record states. Draft or observed material must not silently become confirmed guidance in the destination.
- 3Review the destination
Check its access controls, retention and provider terms. The owner must decide which records may be shared externally.
- 4Choose retrieval or training deliberately
The archive is a portable knowledge corpus, not a ready-made provider fine-tuning dataset. Fine-tuning requires separately prepared and evaluated task-specific examples.
If export cannot finish
- If ownership or the selected scope changes while the ZIP is assembled, refresh and select the scope again.
- If the bounded export limit is exceeded, download smaller project batches. The service rejects oversized exports instead of silently truncating them.
- If a saved encrypted record cannot be decoded, report the error. Do not assume the service skipped it or produced a complete archive.
- Keep a downloaded archive secure. Erasing company memory later does not delete copies you already downloaded or gave to a provider.