Projects, policies and activity
Manage Polaris projects and connections, set how much context agents may take, and see what they asked for.
Everything for Polaris is under Polaris in the app.
Overview and projects
Your connected computers and clients (Claude Code, Codex), and each project with its index status: live, how fresh, how much of the code is covered. A project can also be a remote project from an uploaded ZIP snapshot.
Connections and tokens
Each connection has its own token (oxmcp_…), bound to your organisation and projects, with an expiry. Revoke a connection to cut it off at once.
Policies
Limit what agents may take, for the whole organisation or one project:
| Policy | |
|---|---|
| Profile | How much context a call returns by default, and which profiles are allowed. |
| Tokens | The most a single result may contain. |
| Sources | Which sources may be read. |
| Depth | How far related-code lookups may go. |
| Excluded paths | Paths never returned. |
A project can only narrow its organisation's policy. Every change is a versioned, audited record; Reset inherits again.
Activity
Every call with its tool, status, duration and size, never the code itself. Use it to see how agents use Polaris and spot a misbehaving client.
Optimisation
Polaris V2 adds the live index and the task-context, symbol and related-code tools. Switch it off for a project or the organisation if a client misbehaves; the basic tools keep working.
Evaluation
Compare how an agent performs a task with and without Polaris: tokens, calls and time. Telemetry beyond metadata is opt-in.