Skip to content
Start free

ProductGovernance

Run AI like infrastructure.Budgets, breakers, and a paper trail.

Governance gives owners and admins the controls they expect from production systems: spending limits that hold, providers that fail over cleanly, prompts you can roll back, and an audit log of every change, all in a separate admin console.

Start freeSee pricing
on monthly workspace spend
Soft and hard limits
with fallback routes
Circuit breakers
for every versioned prompt
Prompt rollback
for OpenAI and Anthropic
Your own keys

How it works

Controls you set once,enforced on every call.

  1. 01

    Set the boundaries

    Choose which providers may see project content, add your own API keys, and set monthly soft and hard limits.

  2. 02

    Route with care

    Tasks call aliases, not raw model IDs. Each alias has a primary and a fallback, and breakers stop sending load to a failing route.

  3. 03

    Review and roll back

    Every invocation records provider, model, prompt version, tokens, and cost. Roll a prompt back in one step and see it in the audit log.

Budgets

Spending that stays inside the lines.At every level.

Budgets nest from a system safety cap down to the workspace, project, run, and single task, and a lower level can never exceed what remains above it. Soft limits warn admins. A hard limit refuses new runs with a clear message instead of a surprise invoice.

  • A cost band before expensive runs, the tracked cost after
  • Run modes that trade depth for cost: economy, balanced, deep
  • One usage record for every provider call and paid search
  • Explicit caps on tool calls, concepts per run, and fetched bytes
Visit the trust center

Capabilities

Operate with confidence,not with guesswork.

  • Bring your own keys

    Connect your own OpenAI and Anthropic keys and allow each provider explicitly, after a clear disclosure.

  • Circuit breakers

    Rolling error rates trip a breaker per route. Optional load stops, probes stay sparse, and the console shows the degraded state.

  • Routing aliases

    Tasks name an alias such as deep_generation, so models can change in configuration without touching code.

  • Versioned prompts

    Prompts are versioned, overridable per workspace, and gated by a deterministic eval suite. Roll back in one step.

  • Audit log

    Members, keys, providers, prompts, budgets, and exports. Every change is recorded with who made it and when.

  • Fixture mode for self-hosting

    Self-hosted deployments can run the full pipeline offline with deterministic outputs for demos and tests.

What you getControl without friction.

Admins get a dedicated console. Everyone else gets runs that behave predictably and stay within budget.

  • Monthly soft and hard budget limits
  • Usage and cost records, with no raw prompt logging
  • A provider allowlist with explicit disclosure
  • Breakers and fallbacks for each route
  • Prompt versions with one-step rollback
  • An audit log in a separate admin console

Explore freely,within limits you set.

Explore a sample project first, then connect providers and set budgets when you are ready.

Start freeSee a sample evaluation