EveryAPI handbook
One OpenAI-compatible endpoint, one key, and one bill in front of 40+ upstream AI providers. These pages document the gateway, the CLI, billing, the marketplace, and running your own deployment.
Every page here also ships inside the everyapi binary and reads the same offline — everyapi docs list for this index, everyapi docs <topic> for one page, everyapi docs search <query> across all of them. That is the same text, from one source; this site is generated from it.
What EveryAPI is
EveryAPI is an AI API gateway: one OpenAI-compatible endpoint, one key, and one bill in front of 40+ upstream providers — OpenAI, Anthropic, Google, Azure, AWS Bedrock, DeepSeek, xAI, and the rest. It also runs a marketplace where ordinary users mount their own upstream capacity as sellers, and a BYO-GPU supply side where idle hardware serves open models.
Quickstart
From nothing to a working request. Two paths: the CLI, if you want to point an existing AI coding tool at the gateway, and a raw API key, if you are writing code.
The HTTP API
EveryAPI speaks three request dialects on one host: OpenAI (Chat Completions and Responses), Anthropic Messages, and Google Gemini. Pick whichever your SDK already emits — the gateway converts to the upstream provider's format and converts the response back.
Models, catalogs and routing groups
Which models you can call is a property of your key, not of the platform. The catalog you see is the intersection of what the deployment has channels for, what your routing group covers, and what the key itself is restricted to.
API keys
A relay API key — sk-everyapi-… — is what your code, your SDK, and your AI tools authenticate with. It is separate from your login session: signing out of the CLI does not revoke a key, and a leaked key is revoked on its own.
The everyapi CLI
One binary. It signs you in, launches AI tools through the gateway, and exposes the whole account surface — keys, usage, wallet, marketplace, supply — without opening a browser. It is also the MCP server (everyapi mcp) and the sidecar EveryAPI Connect drives.
Launching AI tools with everyapi use
The main reason to install the CLI. Every AI coding client has its own idea of which environment variable holds the key, whether the base URL wants a /v1 suffix, and where its config file lives. everyapi use knows all of them.
Money: quota, wallet, subscriptions
Every request is settled after the response, from the token usage the upstream actually reported. What it draws from — wallet balance or a subscription — is a preference you set.
The MCP server
The everyapi binary is also a Model Context Protocol server. Run it as everyapi mcp and an AI agent — Claude Code, Cursor, Codex CLI, any MCP client — can check your balance, list seller channels, mount a channel, or withdraw earnings, without you opening a terminal.
The sanitizer proxy
A local HTTP proxy that masks secrets in outbound requests before they reach the gateway, and restores them in the response on the way back. The mapping table exists only in that process's memory: never written to disk, never sent over the wire, dropped on exit.
Driving macOS apps
everyapi computer reads and controls local macOS application windows through Accessibility. It is the one surface in this CLI that touches things outside EveryAPI, so it is also the one with the most explicit boundaries.
EveryAPI Connect (desktop)
The desktop client — Tauri 2 and React, with the EveryAPI Go CLI bundled as a sidecar. It signs in through the same device flow, finds the AI tools installed on the machine, and installs, configures, or launches them without anyone opening a terminal.
The web dashboard
https://app.everyapi.ai — sign-up, keys, spend, and everything that genuinely needs a browser: card payment, passkey registration, email verification, OAuth binding.
Selling capacity on the marketplace
The marketplace lets an ordinary user mount their own upstream channel — a provider API key, or an OAuth-linked subscription — and earn a share of what buyers spend routing through it. The platform takes a configurable cut.
Edge nodes: selling your own GPU
An edge node turns idle hardware into marketplace supply: a small agent plus Ollama, serving open models to buyers routed through the gateway. everyapi edge condenses the deployment into one command set, so nobody has to hand-copy a compose file, fill in a .env, or move a registration token around.
Running your own deployment
EveryAPI is a Go backend you can run yourself. The default image is the API gateway only; an optional image embeds the dashboard for a single-container deployment.
When something is wrong
Symptoms and their causes, in the order you are likely to hit them: sign-in and session errors, gateway 401s and 429s, a model list that is shorter than expected, tool launches that do not find their binary, and the macOS permission grants that are easy to give to the wrong application.