Get from first request to a governed rollout.

A practical guide to testing the workspace, configuring controls, bringing in teams, and verifying results.

Open the architecture overview

Where every part of a request lives.

  1. App

    Familiar chat in the browser

  2. Policy

    Budget and model rules before the call

  3. Model provider

    Receives only the approved request

  4. Storage

    Transcripts and usage in SQLite

  5. Audit

    Hash-chained evidence records

Diagram separating app, policy, storage, and audit in your environment from the external model provider across a boundary. Sensitive values leave as placeholders Restored locally after inbound policy Only the approved request crossed

01

Follow one request end to end

See how an authenticated employee request moves through model policy and the outbound gate to an approved provider, then returns through inbound checks and recorded evidence.

  • Authentication adds tenant, user, and group context
  • Policy determines eligible models and request action before provider access
  • The route, usage, response, and audit evidence stay connected

02

Start with a guided sandbox

Use the seeded workspace to experience employee chat, policy decisions, shared resources, and savings reporting without preparing production data.

  • Try the synthetic secret and restricted-results prompts
  • Trace each resulting policy decision and its audit evidence
  • Review shared resources, routing outcomes, and Finance usage

03

Turn policy into action

Combine detector, direction, scope, and action to make company policy part of each request instead of a separate review process.

  • Evaluate outbound prompts and inbound responses independently
  • Choose allow, flag, redact, warn, block, or require approval; restore valid placeholders locally for the original requester after inbound policy, or use irreversible masking
  • Keep violation and approval work connected to supporting evidence

04

Bring the right people in

Use roles and groups with local accounts, TOTP, Google Workspace or Microsoft Entra ID sign-in, verified-domain sandbox access, and SCIM Bearer operations.

  • Enforce tenant and group boundaries
  • Use local accounts and TOTP or Google Workspace and Microsoft Entra ID sign-in
  • Create, read, deactivate, or delete users and groups through SCIM Bearer

05

Give each team the right models

Start with platform credentials, optionally add BYOK, and approve providers and models by group while Auto adapts routing to each question and shared capacity.

  • Show employees only provider families backed by their enforced allowlist
  • Use adaptive model routing across the approved catalogue
  • Keep platform access ready by default while BYOK remains optional

06

Operate with confidence

Choose managed hosting with an agreed operating scope, or deploy the Private Edition on supported Linux with Docker Compose and Node 24 LTS while our team assists implementation and handover.

  • Define hosting, backup, recovery, upgrade, and support operations
  • For Private Edition, verify the host, HTTPS, persistent data volume, and provider credentials
  • Test health, login, policy outcomes, audit integrity, backup, and recovery before launch
Compare deployment options →

07

Integrate predictably

Use the authenticated API with consistent errors, tenant authorization, and idempotent mutations so integrations can handle success and retry paths reliably.

  • Check sessions and permissions before tenant data access
  • Send a stable Idempotency-Key with documented mutations
  • Use bounded pagination and structured filters for list endpoints

08

Verify the rollout

Use policy, audit, usage, and savings views to confirm that employees have the right access and administrators can explain every recorded outcome.

  • Review allowed, redacted, warned, blocked, and approval-required decisions
  • Compare routing outcomes with the team’s configured benchmark
  • Export evidence and reports for operational review