Team server overview

What it is

The team server is the same memory engine as the local install, run once for many people. It ships in the same package as the tam-team command and as a Docker Compose profile (docker-compose.team.yml).

Employees do not install models or databases. Their AI client connects over HTTPS to the server’s /mcp/ endpoint with a personal token. Models, indexes and data stay on your server.

Three kinds of memory

Scope	Who can read	Who can write	Typical content
Personal	Only the owner	Only the owner	Private notes, investigations, drafts
Team (one per department)	Members of that team	Members with the editor role	Runbooks, conventions, decisions of that team
Shared	Every authenticated user	Every authenticated user	Company-wide onboarding, glossary, policies

• Saving defaults to personal. People choose a team or shared scope explicitly.
• Searching covers every scope you can access at once, or one scope if you pick it.
• Tags organise topics; they never grant access. Access comes only from scope and membership.

What else you get

• Authorship. Every record stores who wrote it and from which client, taken from the token. Nobody can save under someone else’s name.
• Edit history. An edit creates a new revision and keeps the old ones, with author, UTC time, reason and before/after state. expected_revision prevents silently overwriting a colleague’s change.
• Safe retries. Writes accept a request_id (UUID); repeating the same request returns the first result instead of saving twice.
• Isolation. Each personal, team and shared scope is a separate SQLite database with its own graph and index, served by a separate worker process.
• Web interface on the same port for signing in with a token, searching, browsing, saving, editing and viewing history. New in 14.6.0 A full dashboard with roles, reports and onboarding at /dashboard/.

What the remote catalogue contains

Remote clients see 23 tools, not the full local catalogue. The eight memory tools: memory_scopes, memory_save, memory_recall, memory_get, memory_update, memory_delete, memory_history and memory_export. Tools that read files on the server machine are deliberately not exposed. Moving an existing record between scopes is not available yet; saving directly into the chosen scope is.

New in 14.6.0 Plus memory_report for personal, department and company activity reports, and 14 onboarding_* tools for department onboarding.

A person can still run a local TAM for their own machine alongside the team server; the two are independent.

Current limits

• Authentication uses pre-issued personal Bearer tokens. There is no interactive OAuth sign-in.
• Users, teams, memberships and tokens are managed with the local tam-team command on the server, and since 14.6.0 also in the web dashboard.
• Plan at least 4 GiB of RAM for three warm workers; see Install.

Setup in five steps

1. Install the server with Docker Compose or the Python package.
2. Put it behind HTTPS with nginx or Caddy.
3. Create users, teams, memberships and tokens.
4. Connect each employee’s client.
5. Schedule backups and read the security model.