● Documentation · V1 · MAY 2026
Build with deeprelay.
OpenAI-compatible serverless inference (chat, embeddings, image and video) on deeprelay. A REST API, a CLI, and an MCP server for coding agents. These docs cover everything you can do as a developer on the platform.
§ 01 · Start here
JUMP IN
Run your first completion
From zero to a streaming chat completion in under five minutes.
Quick start →REFERENCEEvery /v1 endpoint
Schemas, examples in curl / Python / Node / Go, response envelopes.
Endpoint reference →FOR AGENTSWire up an MCP client
Claude Code, Claude Desktop and Cursor, with spend bounded by a confirmed session budget.
MCP server →ALL TOOLSThe developer hub
CLI, REST API, and MCP in one page: one key, first call in minutes.
Explore the hub →§ 02 · Contents
Getting started
02 chapters
CLI
02 chapters
API
04 chapters
- 03ConventionsAuthentication, errors, pagination, idempotency, and rate limits.
- 04Endpoint referenceEvery /v1 endpoint with curl, Python, Node, and Go samples.
- 05Inference APIOpenAI-compatible chat, embeddings, image generation, and async video jobs.
- 06WebhooksLifecycle events, HMAC signatures, retry policy, verification.
Integrations
04 chapters
- 07OverviewOne base URL and one key: deeprelay inside the SDKs and gateways you already run.
- 07.1Vercel AI SDKgenerateText, streamText, a Next.js useChat route, tools, structured output, and embeddings through the openai-compatible provider.
- 07.2LiteLLMThe Python SDK and the LiteLLM proxy with aliases, fallbacks, and a master key.
- 07.3PortkeyPortkey's gateway in front of deeprelay: saved providers, retries, fallback configs.
SDKs
02 chapters
Guides
07 chapters
- 12Getting startedAccount, credit, and your first inference call in minutes.
- 13Refer a friendShare your invite link: when your friend deposits $50, you each get $25.
- 14Serverless Inference APIThe full /v1 guide: chat, embeddings, image and video, tool calling, the Anthropic Messages endpoint, plan coverage, and errors.
- 15Claude CodePoint Claude Code at deeprelay models with one setup command: what works, what is rejected, how to undo it.
- 17Codex CLIRun Codex CLI on deeprelay models through the OpenAI Responses endpoint: setup, what works, known limitations.
- 16Coding agentsCline, Kilo Code, Roo Code, OpenCode, Aider, Continue, Zed, Cursor, Pi, Kimi Code, Deep Code, DeepSeek Harness and ZCode on deeprelay: base URL, key, model ID, gotchas.
- 18Agent skillInstall the deeprelay skill into Claude Code, Codex or OpenCode and ask your agent about quota, usage and errors in plain English.
Need a hand?
We respond fast. Email support@deeprelay.ai for anything: billing, integration help, weird edge cases.