phaseoteam/phaseo
Build on the Phaseo AI gateway: integration, migration, model selection, SDK and CLI skills, plus the Phaseo MCP for live models, pricing and usage.
Migrates an existing OpenRouter integration to the Phaseo Gateway end to end: usage inventory, endpoint, key and SDK swap, model ID and suffix mapping, provider objects, plugins, presets and OpenRouter-only fields, validation and a migration report. Use when code contains openrouter.ai URLs, sk-or-v1 keys, OPENROUTER_API_KEY, @openrouter packages or OpenRouter provider objects and the user wants a full migration. Not for other source platforms (use phaseo-migration) or new Phaseo integrations (phaseo-gateway).
Uses the connected Phaseo MCP for live data: the interactive model explorer, model and provider details, token cost estimates, credits, activity, analytics and privacy-minimised request health, and reviewed text inference. Use when comparing AI models or provider availability through Phaseo, estimating costs, or investigating a Phaseo account's usage or a request. Also manages API keys, budgets, spend rules, provider keys, guardrails, presets, settings, webhooks and observability destinations when the user connected with account management. Not for writing integration code (use phaseo-gateway or an SDK skill) or billing and payments.
Builds agent loops on the Phaseo Gateway with the TypeScript Agent SDK, @phaseo/agent-sdk: createAgent, defineTool, createGatewayAgentClient, bounded model retries, resumable runs from app-owned state, presets, managed search and concurrent tool calls. Use when a project imports @phaseo/agent-sdk or needs a tool-calling agent, a resumable workflow or a human review step on Phaseo. Not for single requests (use phaseo-sdk-typescript) or agent SDKs in other languages.
Builds Phaseo async video, batch and webhook integrations: creating jobs and persisting their IDs, polling or websocket status helpers, idempotent webhook consumers for the standard payload, and recovery when local state and the gateway disagree. Use when a project generates video, submits batches, receives Phaseo webhooks or has stuck long-running jobs. Not for synchronous text, image or audio requests (use phaseo-gateway or an SDK skill).
Operates Phaseo from the terminal with the official phaseo CLI: install and update, browser or device OAuth login, API keys, workspaces, presets, BYOK, budgets, spend rules, rate limits, notifications, observability, routing, identity and other workspace settings, plus credits and human-confirmed top-ups, analytics and request lookup by ID, with --json output for automation. Use when the user runs or wants phaseo commands, scripts Phaseo admin work or debugs a request from a shell. Not for application code that calls the API (use phaseo-gateway or an SDK skill).
Integrates applications with the Phaseo Gateway over OpenAI-compatible REST, the TypeScript SDK (@phaseo/sdk), the Python SDK (phaseo) or the Vercel AI SDK provider (@phaseo/ai-sdk-provider): base URL, API key auth, attribution headers, model IDs and smoke tests. Use when adding Phaseo to a codebase, wiring a new endpoint, seeing api.phaseo.app or PHASEO_API_KEY, or debugging auth, routing or response-shape errors. Not for moving off another provider or gateway (use phaseo-migration) or choosing models (phaseo-model-selection).
Configures and debugs Phaseo guardrails on API keys: model and provider restrictions, budget limits, prompt-injection handling, sensitive-information detection, preview testing and verification in activity and request logs. Use when a rollout needs to limit what a key can call or spend, or a request was blocked by a guardrail. Not for routing or fallbacks (use phaseo-routing-presets) or reviewing spend after the fact (phaseo-usage-monitoring).
Migrates existing LLM integrations to Phaseo from OpenAI, Anthropic, OpenRouter, Vercel AI Gateway, LiteLLM, Cloudflare AI Gateway, Helicone or LLM Gateway. Swaps base URL and API key with a minimal diff, maps model IDs via GET /v1/models, maps fallbacks, caching, headers and observability, and smoke tests. Use when code has api.openai.com, openrouter.ai, ai-gateway.vercel.sh, gateway.ai.cloudflare.com, helicone.ai, api.llmgateway.io or a LiteLLM proxy, or the user asks to switch to Phaseo. Not for new integrations (phaseo-gateway) or deep OpenRouter work (openrouter-to-phaseo-migration).
Chooses, compares and pins Phaseo model IDs for a workload by modality, context window, token price, capabilities (tools, structured output, reasoning, vision), provider availability, data policy and lifecycle, using GET /v1/models or the Phaseo MCP. Use when someone asks which model to use, the cheapest or best model for a task, whether a model supports a feature, what a workload will cost, or what replaces a deprecated model or alias. Not for wiring requests (phaseo-gateway), routing or fallback policy (phaseo-routing-presets), or tracking spend (phaseo-usage-monitoring).
Configures and debugs Phaseo response healing, the response-healing plugin that repairs malformed structured JSON output, at the workspace, preset or request level, with log-based verification. Use when a structured-output or JSON-mode workflow returns invalid JSON, or healing should be rolled out or ruled out. Not for choosing models that support structured output (use phaseo-model-selection) or validating JSON in application code.
Rolls out and debugs Phaseo presets and routing policy: choosing between a fixed model ID, a preset (@slug) or a router such as phaseo/free, provider narrowing and fallbacks, stable request shapes for prompt caching, and request-log investigation. Use when requests should share defaults, routing picks an unexpected provider, fallbacks misbehave, or someone expects a preset's response caching to work (it is not available yet). Not for choosing which model to use (use phaseo-model-selection) or key restrictions and budgets (phaseo-guardrails).
Builds Phaseo integrations with the official Python SDK, phaseo: one shared client, generate_response and streaming, images, audio, async video, presets, model discovery and request IDs for debugging. Use when a Python project depends on or should adopt phaseo, in scripts, notebooks or services. Not for TypeScript (use phaseo-sdk-typescript) or code that keeps the OpenAI Python SDK with a swapped base URL (phaseo-gateway).
Builds Phaseo integrations with the official TypeScript and JavaScript SDK, @phaseo/sdk: one shared client, generateResponse and streaming, images, audio, async video, presets, model discovery and request IDs for debugging. Use when a project depends on or should adopt @phaseo/sdk in Node.js, Bun or edge runtimes. Not for agent loops (use phaseo-agent-sdk-typescript), Python (phaseo-sdk-python) or code that keeps the OpenAI SDK with a swapped base URL (phaseo-gateway).
Monitors and debugs Phaseo spend and usage through REST, the phaseo CLI or the Phaseo MCP: credit balance, per-request cost and tokens, log and generation lookup by request ID, analytics by day, model, key or label, 429 rate limits, budgets, alerts and observability destinations. Use when asked why a request cost X or failed, where credits went, to export usage, or about x-request-id, /v1/credits, /v1/analytics, /v1/logs or Retry-After. Not for key restrictions or guardrail setup (phaseo-guardrails), choosing models by price (phaseo-model-selection) or first integration (phaseo-gateway).
Configures and debugs web grounding on Phaseo: the gateway:web_search and gateway:web_fetch tools, native provider web search, citations and request-log checks for search-heavy flows. Use when responses need current web information or citations, or search tool calls fail or return nothing. Not for calling your own function tools (use an SDK skill) or building agent loops (phaseo-agent-sdk-typescript).