Integrate with AI

Integrate only Claude models through Essevin in the current project and follow this contract.

Scope
- Implement only what is required for the Claude API or Claude Code. Do not add, configure, or test an unselected model family.
- Preserve the existing framework, SDKs, and code style. Make the smallest changes required for this integration.
- Determine whether this is a local single-user tool, an internal service, or an application serving downstream customers or tenants before choosing key placement and architecture.

Key and model
- Use only a Claude-group key. In a multi-model project, prefer ESSEVIN_CLAUDE_API_KEY or the existing secret manager. Reuse ESSEVIN_API_KEY only in an explicitly single-model project.
- Do not create a key, read another model family key, or ask me to paste a real key into the conversation. Stop and ask me to configure it locally or in the secret manager when unavailable.
- First call GET https://api.essevin.com/v1/models with this key. Select only model IDs beginning with claude- from the response.
- On model_not_found, fetch /v1/models again and use a complete returned model ID. Never guess or rewrite a model name.

Connection
- Use the Anthropic protocol with https://api.essevin.com as the Base URL, without /v1.
- Use api_key with the Anthropic SDK or x-api-key for native HTTP. Existing Anthropic code should keep native authentication.
- The SDK sends Claude API requests to POST /v1/messages. Do not convert them to OpenAI Chat Completions.
- Claude Code uses ANTHROPIC_BASE_URL=https://api.essevin.com and ANTHROPIC_AUTH_TOKEN sourced from the selected Claude key.

Implementation
- Inspect the current dependencies and configuration. Reuse an installed SDK and preserve every unrelated provider configuration.
- Put the Base URL, key variable name, and default model in the existing configuration system. Update .env.example without adding a real key.
- Downstream or multi-tenant calls must keep the key on the server and reuse caller authentication with a model allowlist, request limits, timeouts, rate limits, usage attribution, and redacted logs.
- Do not add a proxy service or multi-tenant gateway to a local single-user tool.
- Choose the default only from Claude models returned by /v1/models. Do not hard-code a documentation example.
- Pass Anthropic SSE through end to end and preserve upstream 4xx or 5xx statuses and safe error details.

Validation and handoff
- Send one minimal real request after configuration. Add a streaming test only when the project actually streams.
- Ask a returned Claude model to reply with only ok through Messages and verify its content.
- Report changed files, the key variable name, model ID, endpoint, HTTP status, and a redacted response summary. Include start, verification, and rollback commands.