3M
free tokens / day
Open models.
Decentralized GPUs.
One API.
Inference for AI agents on open-weight models, served by the GPUs of independent miners in the AETRON network. OpenAI and Anthropic request shapes, tool calling and streaming, 3M tokens a day free.
—
per 1M tokens after the free pool
1
service: one key and one bill for the whole catalog
⌘
One service
One key, one base URL and one bill for every model in the catalog, instead of a provider per model. Switch models with a string.
∞
Free every day
3M tokens a day, no card. Paid models draw on prepaid credit only after the free pool is spent.
⇄
Stays answered
If a node is unavailable, the router retries on another provider or walks your fallback list. You get an answer, not a 502.
Built for long-running agent tasks
On a long task an agent makes hundreds of calls in a row. The router gives it the primitives and puts a fence around it.
Tool calling
OpenAI and Anthropic tool shapes. Calls arrive as tool_calls, results go back as tool messages, in a stream too.
Streaming
Server-sent events with usage in the final frame, so an agent shows progress and counts its tokens.
Two API dialects
Frameworks written for OpenAI and frameworks written for Anthropic both work: same key, same base URL.
Budgets per key
Daily token, request and USD caps on each key. A looping agent stops at the fence, not at your balance.
Guardrails
Prompt-injection protection, PII filters and model allowlists attached to the key the agent uses.
Request log
Every call with model, tokens, cost and latency, so you see what the agent actually did.
AETRON Code
An agentic coding client for the AETRON Router. It reads and edits your code, runs tools, and works with any model from the catalog, on your machine with your API key.
- Agent loop with tools and code edits
- Skills and MCP servers
- Any catalog model, free ones too
Download for macOSmacOS · universal · .dmg · 195 MBDownload for WindowsWindows · x64 · installer · 109 MBv0.1.1
The network
What actually answers your request
AETRON is a decentralized network of GPU miners running open models. The router is the front door: one key, one URL, and the network does the work.
Open the explorer →Independent miners
Models run on GPUs of independent operators, not in one data centre. Capacity grows as miners join.
Models live in neuronets
A neuronet is a group of miners serving a set of models, hundreds in one. The router knows which neuronet serves which model and picks a live provider for each request.
Open ledger
Neuronets, miners and blocks are public on the AETRON chain. Watch the network live.
Featured modelsBrowse all 3 models →
[ 01 ]
Create an account
Sign up in seconds. You get 3M free tokens a day, right away — no card.
[ 02 ]
Make a key
Generate an sk-aetr key. The secret is shown once; only its hash is stored.
[ 03 ]
Point your client
Set the base URL and key in any OpenAI or Anthropic SDK. That’s the whole integration.
Works with
Agent frameworks and clients, unchanged
Set the base URL and the key. Cursor, Cline, Continue, Aider, LangChain, the Vercel AI SDK, LiteLLM, AETRON Code — anything that takes an OpenAI or Anthropic base URL.
Docs →curl https://api-router.aetron.io/v1/chat/completions \ -H "Authorization: Bearer $AETRON_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "aetron/qwen3.5-9b-gguf-q4-k-m", "messages": [{ "role": "user", "content": "hi" }] }'