Introduction
One API for configured AI models, with organization governance, usage visibility and prepaid credit.
Tokamak gives developers and organizations one place to connect applications to AI models, manage access, understand usage and control spending. Your application or coding tool sends requests to Tokamak; Tokamak authenticates the caller, applies the relevant limits and forwards the request to a configured upstream provider.
Use the customer workspace to connect your first application, invite colleagues, group work into teams and manage budgets. Platform operators manage providers, the model catalog and deployment settings in a separate administration console.
Who Tokamak helps
| Audience | What you can do |
|---|---|
| Developer or individual user | Create a key, connect an SDK or coding tool, and inspect your requests and applicable limits. |
| Organization owner or admin | Create an organization, invite people, manage teams and set organization, team or member budgets. |
| Billing administrator | Review organization credit and usage, set budgets, buy credit and manage payment settings when enabled. |
| Platform operator | Configure providers, models, pricing and deployment services, with explicit separation from tenant administration. |
What you get
- A common API entry point. Use OpenAI Chat Completions, Anthropic Messages or OpenAI Responses with a model and dialect supported by your deployment.
- Provider credentials kept on the server. Applications use Tokamak credentials instead of receiving upstream secrets.
- Separate organizations and teams. Keep membership, billing and governance in the correct organization; attribute keys to teams when needed.
- Usage and spending visibility. Explore request cost, tokens, reliability and individual requests with the permitted audience and time range.
- Coding-tool sessions. See who is coding through Tokamak right now, when your teams work and, with prompt summaries on, what they worked on, when a platform administrator switches Insights on for your organization.
- Automatic model choice. Name
tokamak/autoand each request goes to the cheapest model of your policy that answers it well, when your deployment enables auto routing. - Your own provider accounts. Route Anthropic and OpenAI traffic through your organization's own keys, metered at your price and not charged to your credit, when your deployment enables bring your own key.
- Budgets and prepaid credit. Apply usage caps separately from the funds available to pay for requests.
- Your existing tools. Connect an SDK directly or use the Tokamak CLI with supported coding tools.
Available models, login methods, payment options and service terms depend on the deployment. These docs do not promise a particular model catalog, price, free-credit allowance or service-level agreement.
How it works
A gateway authenticates and authorizes the request. Core selects the upstream provider, applies usage limits and coordinates credit admission and settlement when billing is enforced. Auth owns identities, sessions, keys and permissions. Billing owns prepaid wallets, holds, payments and the ledger.
Tokamak forwards provider-native requests rather than translating between dialects. It substitutes the upstream model identifier and requests usage details for streaming Chat Completions. Client tool calls remain the client's responsibility; provider-hosted tools remain the provider's.
Start here
- Follow the quick start to connect your first application.
- Tour the workspace to understand Home, keys, usage, budgets and credit.
- Set up your organization and teams when you are ready to collaborate.
- Read the customer FAQ before rolling Tokamak out to colleagues.
For a technical introduction, see Core concepts. To run your own deployment, start with Self-hosting.