Home / Features / Privacy and local-first

Local-first, private, and offline-capable AI with osFoundry

Run models on-device, keep private channels off the cloud, and work offline. Your data, your machine.

osFoundry is local-first: a desktop app that runs models on-device, keeps private channels from ever syncing, and works fully offline. End-to-end encryption protects what does sync. The local sync engine projects your workspace into your file system, scoring and indexing happen device-side via osIndex, and a no-leak local-LLM mode guarantees no prompt ever leaves the machine.

Quick answer

Local-first desktop app — works offline, syncs when online.
Private channels never leave your device, period.
End-to-end encryption for everything that does sync.
No-leak local LLM mode — provider-neutral, never any cloud routing.
Local sync engine — workspace as a folder on your disk.

What it is

Most AI products are cloud services with a desktop shell: your prompts, files, and history live on someone else’s servers, and “privacy” means a policy promise. osFoundry inverts that. The desktop app is the primary home of your workspace — the local sync engine projects it into your file system as ordinary folders and files, models can run entirely on-device, and cloud features are opt-in additions rather than the foundation.

Privacy is enforced at the architecture level, not the policy level. A private channel is excluded from the sync queue at creation — its messages and attachments are never uploaded, so there is nothing on a server to leak, subpoena, or misconfigure. The no-leak local-LLM mode applies the same idea to inference: it is a workspace-level switch that disables every cloud route, so no prompt can leave the machine regardless of which chat or agent sends it.

Everything that does sync — because you chose multi-device or team features — travels end-to-end encrypted with workspace-controlled keys. The result is a gradient you control: fully offline and local, selectively synced, or cloud-augmented, per workspace and per channel.

Key capabilities

Local-first sync architecture — bidirectional, conflict-aware.
Private workspace mode — flagged data never leaves the device.
Offline-capable apps — Room Apps work without network.
Device-side osIndex scoring — model scoring runs on your hardware.
No-leak LLM mode — workspace setting that disables all cloud routes.
End-to-end encryption with workspace-controlled keys.

How to do it in osFoundry

Flag what is private — Create a private channel for sensitive work. It is excluded from sync at creation — messages and attachments are never enqueued for upload, and other devices on your account cannot see it.
Go local for inference — Install a local model from the catalog, then enable no-leak LLM mode in workspace settings to disable all cloud routes for prompts.
Verify it yourself — Disconnect from the network and keep working — chat with local models, edit notes, run offline-capable Room Apps. Sync resumes for non-private data when you reconnect.

Use cases

Legal and back-office teams: Draft against contracts and HR documents with a local model in a private channel — nothing ever leaves the machines that hold the files.
Consultants under NDA: One private channel per client. Client material stays on-device while general research uses cloud models in normal channels.
Frequent travelers: Full workspace on a laptop: local models, notes, and apps work in flight; encrypted sync catches up on landing.

Frequently asked questions

Does osFoundry require a cloud account to use?

No. The desktop app runs locally with no sign-in required. Cloud features (sync, hosted GPU, public app URLs) are opt-in.

Can I work entirely offline?

Yes — install local models, disable cloud routes, and the app works without network access. Sync resumes when you reconnect.

How is "private channel" different from a normal channel?

A private channel is flagged at creation. The local sync engine never enqueues its messages or attachments for upload. Even other devices on your account can’t see it.

Which data leaves my machine in the default setup?

Only what you opt into. Private channels never sync. Non-private data syncs end-to-end encrypted when you enable multi-device or team features; local-model prompts never leave the device, and no-leak mode extends that guarantee to the whole workspace.

Can I mix cloud models with local privacy?

Yes — route sensitive prompts to local models (or keep them in private channels) while public-information prompts use cloud models. The routing is configurable per path in osStudio.