Choose approved projects, control each worker’s availability and return checked results to the project owner. Local network operation needs no cloud relay. This is a Preview: automated local peer checks do not establish completed twelve-machine hardware acceptance or a native Windows test.
Find Fleet Mode in Advanced settings
In the build 26 preview navigation, open Central Brain Settings → Advanced → Fleet Mode. Share Brain includes a shortcut. Check the installed release and current download options for your platform; availability varies by platform and release.
Keep the project owner in control
- Reuse trusted pairings; each worker explicitly accepts resource sharing.
- Select approved coordinator-owned projects and supported document-preparation jobs.
- Send only inputs needed for the assigned task into temporary worker storage.
- Check returned results against the job and current asset versions before updating the owner project.
- Inspect progress or pause workers while the owner retains project control.
What the first preview processes
Only supported document-preparation tasks are delegated in this preview. Private profile keys and provider credentials are not transferred.
| Boundary | First preview scope |
|---|---|
| Task | Document preparation for coordinator-owned projects. |
| Inputs | TXT, Markdown, PDF, DOCX, CSV and XLSX. |
| Per-file input limit | 64 MiB. |
| Per-file prepared-output limit | 32 MiB. |
| Indexing and embeddings | Remain on the owner. |
| Redaction-protected projects | Remain local to the owner. |
Trust workers with the actual inputs they process
Local network operation uses direct authenticated encrypted connections. Workers must be trusted because processing permission includes access to the actual files needed for their assigned tasks. Joining a fleet does not share a worker’s own library.
Remote networks need an existing reachable route, such as a VPN. Nearby discovery is not guaranteed across VPNs. Local pairing and fleet operation require no cloud account.
Common questions
Has a twelve-computer fleet been physically validated?
Fleet Mode is designed for up to twelve computers. Physical lab acceptance remains pending; automated local peer checks are not a completed twelve-machine hardware or native Windows test.
Does Fleet Mode distribute embeddings or model inference?
The first preview delegates supported document preparation. Final indexing and embeddings remain on the owner; arbitrary model processing or GPU pooling is not advertised.
Can redaction-protected projects be sent to workers?
No. Redaction-protected projects remain local to the owner in the first preview.
Does joining a fleet share the worker’s own library?
No. Worker resource permission grants access to assigned task inputs, not automatic sharing of the worker’s own library.