Frequently Asked Questions

Till is an API key proxy that creates scoped credentials for AI agents. Activation limits are enforced before requests; token and spend budgets use usage reported by providers and apply to subsequent requests.

Till supports 12 AI providers: OpenAI, Anthropic, Google (Gemini), OpenRouter, Mistral, Groq, Together AI, Fireworks, Perplexity, DeepSeek, xAI, and Cohere. Compatibility and usage reporting vary by provider and endpoint.

For OpenAI-compatible clients, set the base URL to https://api.till.ac/v1 and use the scoped key in the Authorization header. Other providers retain their provider-specific paths; see the proxy documentation.

You can set three types of limits:

1. Activation limits - Cap the number of API calls (50 means exactly 50 requests)
2. Token budgets - Track total input + output tokens reported by the provider
3. Spend budgets - Estimate spend for models present in Till's pricing table

Till reserves activation and a conservative token or spend bound before forwarding, then settles reserved capacity to provider-reported usage. A budgeted generation request must declare max_tokens, max_completion_tokens, max_output_tokens, or Google generationConfig.maxOutputTokens. Spend-limited requests with unknown model pricing are rejected.

Till rejects a request when its conservative reservation would exceed remaining activation, token, or spend capacity. Successful requests settle the reservation to provider-reported actual usage. Provider-side budgets and invoice monitoring are still recommended.

Yes. You can revoke any scoped key at any time through the dashboard or API. Once revoked, the key immediately stops working even if it hasn't reached its limits.

Provider API keys are AES-256-GCM encrypted at rest with tenant/provider binding. Scoped tokens reference your account's connections without revealing credentials. Till decrypts only the selected connection in service memory while proxying. See the security overview.

Till is a proxy control layer, not a general-purpose vault. It stores encrypted provider connections so administrators can rotate a provider once while scoped keys remain provider-agnostic. The database, encryption key, scoped tokens, and running proxy still need normal secret-management controls.

Revoke the lost scoped token immediately. New automatic scoped tokens do not contain provider credentials, so you can issue a replacement without rotating every provider connection unless broader compromise is suspected.

Till's current controlled-beta platform plans are:

Free ($0) - 3 scoped keys, 1,000 activations/month
Pro ($59/month) - 25 keys, 25,000 activations/month
Scale ($249/month) - 100 keys, 250,000 activations/month

These prices apply to new subscriptions. Eligible paid beta subscriptions keep their existing founding price under Till's published price-protection policy. Provider inference is billed separately under your own provider accounts. New accounts are manually onboarded, and existing account holders can authenticate in the dashboard to upgrade.

Till forwards streaming responses. Usage tracking depends on the provider including usage data in the stream; if it does not, token and spend counters may not update. Test the exact provider, model, and SDK combination before relying on a budget.

Till is currently offered as a hosted service. A supported public self-hosting distribution is not currently published.