Questions, answered.
Everything operators and developers ask before running a node or wiring the gateway into an app. Still stuck? Talk to us.
Do I need to know Solana to run a node? +
No. You need a wallet to receive payouts, but the node software handles signing, registry, and settlement. The quick-start script provisions a testnet wallet for development.
Is SolanaLM OpenAI API compatible? +
Yes. The gateway exposes /v1/chat/completions and /v1/completions matching OpenAI’s schema. Point the OpenAI SDK at the gateway URL and pass your Solana wallet address as the api_key.
What models can I serve? +
Anything PyTorch/Transformers supports, anything llama.cpp can load (GGUF), or anything reachable through OpenAI/Anthropic/Cohere/Ollama backends. You declare available models in node config.
How does privacy-preserving inference work? +
You can route a request through a 3-hop onion circuit so no single node sees both the prompt source and the prompt content. Combined with differential privacy on federated rounds, sensitive data stops leaking.
What does it cost to run the gateway? +
The software is MIT-licensed and free to self-host. Your costs are infrastructure: a small VM for the gateway, plus GPUs for any local inference nodes you operate.
How are payments enforced? +
The gateway requires a signed Solana transaction (or a pre-funded session) before forwarding a request to a node. Settlement happens per-request; disputes use on-chain audit logs.
Can I run only training nodes, not inference? +
Yes. Nodes declare their role at startup — inference, training, or proxy — and the registry routes work accordingly. Pure training nodes earn from federated round participation.
Is the federated learning implementation actually private? +
SolanaLM ships secure aggregation and (epsilon, delta)-differential privacy with configurable noise. The defaults give meaningful privacy; the algorithms are tunable for stricter regimes.
Can an AI agent pay for inference on its own? +
That is the design target. Because the gateway takes a Solana wallet as the API key and settles per request in SOL, an agent with its own wallet can fund a session and pay as it calls — no human, no monthly invoice, no seat license. Per-request settlement (rather than per-hour rental) is what makes low-margin, high-volume agent workloads economical.
Why Solana instead of an L2 or another chain for AI payments? +
Machine-payable inference needs finality faster than a human reads the first token and fees below the marginal cost of a call. Solana gives ~400ms finality and ~$0.00025 fees, so settling a payment per request is economically real rather than a rounding error swallowed by gas. Slower or pricier chains force you to batch so aggressively that per-request settlement stops being per-request.
How does SolanaLM fit the DePIN GPU compute trend? +
SolanaLM is the serving-and-settlement layer that generic GPU marketplaces lack. Nodes are containers that register on their first heartbeat, so they run directly on DePIN networks (io.net, Akash, Render, Nosana) or marketplace/spot GPUs (Vast.ai, RunPod). Operators bring the cheapest idle hardware; SolanaLM gives it an OpenAI-compatible endpoint, routing, and per-request SOL revenue.
Is decentralized inference verifiable — can I trust the result? +
Today, every request and payout is anchored to an on-chain audit trail, so you can reconcile what was served against what was paid. Cryptographic verification of the compute itself is on the roadmap, cheapest first: signed attestation of which model and config produced a result, then sampled re-execution that slashes a node reputation on mismatch, and eventually TEE/zkML proofs behind a trust tier for high-assurance workloads. We are honest that full verifiable inference is not shipped yet.
What does it cost to run a production node? +
The smallest credible unit is a single 24GB consumer/prosumer GPU (RTX 4090-class) on marketplace or DePIN spot pricing, serving a quantized 7B–8B instruct model via the llama.cpp GGUF backend — a few dollars a day of hardware against per-request SOL revenue. You develop on Solana devnet for free and go to mainnet with a dollar-scale SOL float, since fees are sub-cent.
Ready to run it?
Clone the repo and send your first request in minutes.