In development

Private AI search for the whole organisation.

Garage Enterprise is Garage for teams. It indexes your organisation's documents, code and, where an administrator allows it, mail, and serves them to AI assistants over MCP. Your company's data remains in the company, and agents are only able to search for what you permit.

Describe Your Team's Needs Explore Garage Today

The guarantee

Two assurances, each one enforced by the product.

Garage Enterprise makes exactly two promises, and every one of them traces to a mechanism an administrator can check.

Your company's data remains in the company.

Documents, code, the index and everything derived from them (chunks, vectors and facts) live only on computers and servers your organisation runs. Nothing goes to Garage, and nothing goes to a cloud AI provider.

The agent on each computer talks to one destination beyond itself: your organisation's enrolled cluster, with its certificate authority pinned.
No cloud AI SDK exists in the codebase, and a test fails the build if one appears.
Answers are generated on the asking person's own computer or on your organisation's own pool, never on a cloud provider.
Every destination is one an administrator can list and audit.

Agents are only able to search for what you permit.

An assistant connected over MCP, a script or a person at the console can search only the documents its user can access, under your organisation's own rules: your directory's users and groups, and the permissions on the share, site or repository each document came from.

Row-level security in Postgres filters every read by the caller's access, so a missing filter in a query cannot leak.
A second check and an audit record on every request.
Administrators choose which sensitivities an assistant may see at all.
View as User lets an administrator check the outcome as any person.

How it works

Garage on every computer, one cluster your organisation runs.

Garage Enterprise keeps Garage's own pipeline, unforked, and adds the pieces a team needs around it.

1
Agents on your team's computers

The Garage Enterprise agent runs on macOS, Windows and Linux. As a collector it indexes what only that computer can read; as a worker it computes embeddings and facts for the pool, only when the computer is idle, on power and allowed to.
2
A cluster you operate

One service on Linux, over sharded Postgres with pgvector, holds the index and enforces access. Agents connect to it outbound only, over mutually authenticated TLS. Run it on its own or as a WAR in your application server.
3
Search your assistants can use

Assistants connect over MCP with your identity provider's sign-in. Hybrid search (vector and full text, fused) runs under the person's access, and every result is checked again before it is returned.

Collections

A collection is a set of documents with one access policy: a person's own folders, a file share, a site or a repository. It is the unit of placement, access, retention and deletion.

Verified work

Two computers with different owners compute each unit of work until a computer has earned trust. Embeddings must agree, facts must be found in the document's own text, and a computer that disagrees is quarantined and its results redone.

Communications stay out by default

Mail and messages are not collected unless an administrator enables them for a source, and that choice is recorded. View as User shows another person's communications only as a count.

For administrators

Check what each person actually gets.

Access comes from your identity provider (OIDC sign-in, SCIM 2.0 provisioning) and from the permissions on each source. Garage Enterprise is private by default: a document reaches only the people its source lets read it.

View as User shows an administrator exactly what a person would see, using the same access check their own searches use. It substitutes that person's rights for the administrator's and never adds to them. It is read-only, needs a reason, ends after 15 minutes and is recorded for audit. Results the person cannot access appear only as a count, never as a title.

Viewing as Jamie Lee Reason: access review · 14 min left

quarterly hiring plan

Engineering hiring plan, Q1 Engineering share · document

Interview loop guide People wiki · document

2 results withheld. Jamie isn't in Finance and Legal.

An illustration of the planned console. Teal marks a borrowed, read-only view.

Status

Where Garage Enterprise stands.

In development

Garage Enterprise runs Garage's open-source pipeline unchanged and adds the cluster, the agent and the access rules around it. The agent's enrollment and work pool, verification of results, directory sign-in and provisioning, and View as User's rules are built; the access-controlled index and search over MCP come next. Features on this page that are not yet built are described as plans.

Help shape Garage Enterprise.

If your organisation wants private, local retrieval for its AI assistants, describe how your team works and what it needs.

Get in Touch