Unified HTTP access
Consumers call Fasten Share over HTTP while the service routes requests and streaming responses to producers.
Open source · Cross-platform · Privacy-first
Share AI compute and quota over one network
Discover online model producers through one service with standard HTTP, streaming responses, and server-side settlement.
Key features
Local & online multi-model
Connect local Ollama, LM Studio, vLLM, or online OpenAI / Anthropic-compatible endpoints (URL + APIKey) into one unified network.
Producer + consumer in one
A single cross-platform desktop app with both roles built in: share your own model compute, or borrow others’ to run inference.
Health checks, auto-delisting
Producers are health-checked on a timer with quota probing; an unavailable endpoint or exhausted tokens are delisted automatically, so consumers only ever see live nodes.
Smart discovery & sorting
Match candidate producers by model protocol or keyword. Default-sorted by latency ascending (fastest first), with an optional online-duration sort.
Multi-instance availability
Producers only need an outbound connection, while Redis Streams coordinates multiple service instances.
How it works
As a producer
- Enter a local URL, or an online URL + APIKey + model ID in the desktop app.
- Pick the model protocol, start sharing — it goes online and runs periodic health checks.
- Consumers on the network can connect directly to borrow your compute / quota.
As a consumer
- Search the network for available producers by model protocol or keyword.
- Pick a producer and generate its server-side endpoint.
- Run inference like a local endpoint, with streaming token responses.
Privacy-first
The upstream API key stays on the producer device and is never sent to the service or consumer.
FAQ
Will my API key be uploaded?
No. The producer injects upstream credentials locally, so they remain on the producer machine.
Do business tokens pass through the server?
Yes. The service performs routing, authoritative metering, and settlement; always use the trusted HTTPS endpoint.
Which platforms are supported?
The desktop app supports Windows, macOS, and Linux, with both producer and consumer roles in one program.
Which model endpoints are supported?
Local Ollama / LM Studio / vLLM, plus online OpenAI / Anthropic-compatible endpoints.