ai-stats-gateway

name

ai-stats-gateway

description

Implement and migrate AI integrations to AI Stats Gateway via OpenAI-compatible REST, the official TypeScript SDK (@ai-stats/sdk), Python SDK (ai-stats-py-sdk), and Vercel AI SDK provider (@ai-stats/ai-sdk-provider). Use when wiring new API calls, replacing OpenAI/Anthropic/OpenRouter/Vercel Gateway base URLs, adding app attribution headers, selecting model IDs, or troubleshooting auth/routing/response-shape issues.

AI Stats Gateway

Use this skill to make high-confidence AI Stats integration changes with minimal diff size.

Workflow

1. Identify integration surface before editing:

• OpenAI-compatible REST calls
• OpenAI SDK / Anthropic SDK pointed at AI Stats
• Official AI Stats SDK (TS/Python)
• Vercel AI SDK provider (@ai-stats/ai-sdk-provider)

2. Select the minimal migration path:

• Keep existing SDK and only swap baseURL + API key when possible.
• Move to official AI Stats SDK only when explicit capability gaps or ergonomics need it.

3. Apply standard AI Stats request contract:

• Base URL: https://api.phaseo.app
• Auth header: Authorization: Bearer <AI_STATS_API_KEY>
• Optional attribution headers: x-title and http-referer
• Discover model IDs from GET /v1/models rather than hard-coding stale IDs

4. Validate quickly:

• GET /v1/health for gateway reachability
• GET /v1/models for model availability/capabilities
• One smoke generation request on the target endpoint

5. Harden before finishing:

• Move keys into env vars if found inline.
• Keep API keys out of commits, logs, screenshots, and client-side bundles.
• Add retries/backoff for 429 and 5xx responses.

Integration Playbooks

• OpenAI-compatible REST: read references/rest.md
• TypeScript SDK: read references/sdk-typescript.md
• Python SDK: read references/sdk-python.md
• Migration patterns (OpenAI, Anthropic, OpenRouter, Vercel AI Gateway): read references/migrations.md

Execution Rules

• Prefer minimal and reversible changes.
• Preserve response shapes expected by callers unless explicitly requested to refactor.
• If model IDs differ between platforms, map through /v1/models and keep compatibility aliases where needed.
• If the user asks for best-effort migration, prioritize working endpoint and auth first, then feature parity.