COMPUT Quick Start

COMPUT is the efficiency layer for AI compute. Connect a provider, send requests through one gateway, and inspect recorded usage.

Choose your environment

The local port is our test configuration, not a universal default. Use your instance's actual port. Local and hosted keys, connections, and usage are separate. For the hosted gateway, create an account and verify your email before signing in.

Connect a provider

In your hosted workspace, connect your own API key for Groq, OpenAI, Anthropic, Google Gemini, Mistral, xAI, DeepSeek, or OpenRouter. The provider key is validated and stored encrypted. The model picker lists models returned by connected providers. The provider directory also lets you browse additional model families available via OpenRouter; browsing them does not connect those providers directly or guarantee model access. Each provider row links to that provider's API-key page. For example, the COMPUT model ID for Groq's GPT-OSS 20B is groq:openai/gpt-oss-20b. Availability, billing, and quotas depend on your provider account. See the provider guide.

Create a client key

In your workspace, create a named COMPUT client key. Store it securely. This key authorizes your application to call COMPUT. Your provider secret belongs only in the provider connection, not the playground client-key field. Do not commit keys or paste them in chat.

Send a request

In Playground, select an active COMPUT key and a connected model, then enter a short non-sensitive prompt. You do not need to paste the key back into its own Playground. After a successful response, copy the cURL, JavaScript, or Python example to use in your application.

For a terminal test, set COMPUT_API_KEY securely in your environment first. This is a client-side example variable, not a server configuration setting.

export COMPUT_BASE_URL="https://api.comput.cloud/v1"
curl --fail-with-body "$COMPUT_BASE_URL/chat/completions" \
  -H "Authorization: Bearer $COMPUT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"groq:openai/gpt-oss-20b","messages":[{"role":"user","content":"Reply with a short hello."}],"max_tokens":128,"stream":false}'

OpenAI-compatible clients use this base URL, COMPUT key, and explicit model ID. Prefix models with the provider ID shown in the picker, such as groq:, anthropic:, google:, xai:, deepseek:, or openrouter:. Older unprefixed Groq model IDs still work. Text-chat streaming is supported. You can opt in to fallback by supplying an ordered models array of up to three models from providers you connected. Fallback is attempted only for a timeout, rate limit, or provider error before streaming response headers are returned; each attempted provider may still charge you. Tools are not supported by the Firestore workspace gateway yet.

Usage and cost

• HTTP 200 and response content confirm a successful request.
• Token counts appear when reported by the provider. Missing usage is not zero.
• Inspect History and Overview for request metadata and token usage.
• Costs & billing can estimate spend when token counts and model rates are available. Set rates from your provider's pricing page for unpriced models. Estimates and forecasts are not invoices or proven savings.
• Set a daily request limit in Settings if you want a request ceiling.
• Provider billing is separate and depends on your own provider account.

Troubleshooting

OpenCode's free tier rejected external gateway use in our test. It is not a supported COMPUT free-trial path. Do not bypass provider restrictions.

Deployment and attribution

Use Saminu/COMPUT and its deployment configuration, not an assumed npm package named comput. COMPUT builds on OmniRoute. Upstream identifiers and attribution remain where technically required. Extended reference pages may describe upstream features that need separate configuration and validation.