Token2Token Client

Share GPU. Earn Indigo. Run any model.

Token2Token Client is the open-source provider software for the Token2Token inference network. It connects idle GPUs and local model engines to a unified marketplace while giving providers control over models, prices, resource use, and monthly Indigo earnings.

Capabilities

• Connect Ollama, LM Studio, llama.cpp, vLLM, and approved OpenAI-compatible upstream APIs.
• Discover installed models and publish them to the live marketplace.
• Configure input and output price per million tokens.
• Select static ready-model or dynamic approved-model-pool operation.
• Set a monthly Indigo earnings limit.
• Connect outbound over an authenticated encrypted relay; no inbound router port.
• Report minimal health, capacity, engine, model, and latency telemetry.
• Run supported managed engines in restricted processes or containers.

Privacy boundary

Ordinary community hardware is not confidential compute. TLS protects traffic in transit and managed runtimes reduce accidental exposure, but a machine owner may still access plaintext in host memory. Do not send sensitive data to Community providers. A future confidential tier will require compatible TEE hardware, confidential VMs, and verified remote attestation.

Repository layout

• apps/desktop — Tauri desktop interface.
• crates/daemon — background service and lifecycle management.
• crates/connectors — local engine adapters.
• crates/protocol — generated and hand-written relay protocol bindings.
• engines/vllm — managed Linux/NVIDIA vLLM packaging.
• docs — provider and security documentation.

Public provider contracts live in the Token2Token Provider Specification. The hosted platform is maintained separately.

Quick start

cargo build --release -p token2token-daemon
./target/release/token2token init \
  --token node_your_enrollment_token \
  --accept-commercial-terms
./target/release/token2token discover
./target/release/token2token run

The desktop app exposes the same controls and packages the daemon as a sidecar. Every tagged release publishes both a standalone CLI and a GUI: AppImage/deb for Ubuntu, an installer for Windows, and a DMG for Apple Silicon macOS.

Developer API keys for upstreams such as DeepSeek and Kimi Open Platform remain on the provider machine. These are ordinary community endpoints—not confidential compute. Consumer OAuth or personal subscription keys must never be published or resold. See Third-party upstream providers.

License

Apache License 2.0. See LICENSE and NOTICE.