Aile
Call premium AI models. Share idle API keys.
Aile (pronounced isle) is a marketplace for AI model endpoints. On one side, sharers list capacity they already have: a machine running a model, a provider subscription, an API key sitting unused. On the other, callers reach all of it through a single OpenAI- and Anthropic-compatible API and pay per call in USDC on Solana (or on Base, where the deployment enables it).
The structural difference from an ordinary API gateway is where the request ends up: the egress is someone else's machine. There is no central pool of capacity to sell, because Aile owns none.
In a hurry?
Jump to the Quickstart: it has a working request for callers and a five-minute setup for sharers.
For callers
You get one base URL, one key, and every model any sharer can currently serve.
1curl https://api.aile.sh/v1/chat/completions \2 -H "Authorization: Bearer sk-aile-..." \3 -H "Content-Type: application/json" \4 -d '{5 "model": "cc/claude-sonnet-5",6 "messages": [{ "role": "user", "content": "Say hello." }]7 }'
Because the wire format is exactly OpenAI's, any SDK works by changing the base
URL. Nothing else in your code moves. POST /v1/messages speaks Anthropic's
format for the same catalogue. For Claude Code, Codex and other coding tools,
npx aile.sh setup does the configuring. Images, speech, transcription, video,
music, embeddings, rerank, web search and scrape have their own endpoints under
the same key, and an agent can buy any of it over MCP.
- Your first request: curl, the OpenAI SDK, the Anthropic SDK
- Model names: how
cc/claude-sonnet-5resolves, and why the prefix matters - Choosing a sharer: pin a provider, cap a price, require verification
- Pay with a wallet: no account at all, where the deployment enables it
For sharers
Install the CLI, connect something you already pay for, test it, and set what you charge.
1npm install -g aile.sh2aile login3aile connect4aile test 1 # a model is listed only after it passes a test
What you connect can be any of three things, and the kind decides who can read your prompts and where the request physically goes.
| Kind | What it is | Who runs the model |
|---|---|---|
| Self-hosted | Ollama, vLLM, LM Studio, llama.cpp on your machine (aile local setup installs one) | You. Weights never leave. |
| Subscription | A linked provider account (API key or consumer OAuth) | The provider. Aile injects the credential. |
| Nodeless | A linked account served with no machine of yours in the path | The provider. Aile dials it directly. |
- Install the CLI and connect an account
- Your rates: a margin from free to list price, overall or per model
- What Aile can see: read this before you link a credential
- Getting paid: settlement, wallets, and the platform fee
What "verified" does and does not mean
For some providers (Codex today), Aile can prove the account link: the provider signs a token carrying a server nonce, which turns "a plausible credential" into "this sharer authenticated with the provider just now." Aile can also show that a credential worked recently, and a model is listed only after it has answered a test request.
It cannot prove what a node serves: model identity, quality, or a silent downgrade to a cheaper model. Advertised connections are claims rather than verified facts. The verification page is explicit about where the line falls.
Where to go next
| If you want to… | Start at |
|---|---|
| Send a request in the next five minutes | Quickstart |
| Understand what you are paying for | Pricing & fees |
| See every endpoint, live | API reference |
| Run every CLI command | CLI commands |
| Know what happens to a prompt | How a request is served |
| Connect an agent | Connect over MCP |
| Fix an error | Troubleshooting |