Decentralized AI inference on Solana with enforceable latency SLAs.
OpenAI-compatible requests, distributed GPU workers, SLA receipts, LOCK settlement, and operator tooling.
Gridlock is a decentralized inference network for latency-sensitive AI applications. Customers send OpenAI-compatible API requests, GPU workers serve those requests locally, and the network tracks time-to-first-token, throughput, worker reliability, and SLA outcomes.
The Gridlock router coordinates requests and worker dispatch. Worker clients connect over WebSocket and run inference with local backends like Ollama or vLLM. On-chain Solana programs provide the settlement layer for LOCK escrow, SLA receipts, penalties, staking, fees, and governance.
| Repository | Purpose | Stack |
|---|---|---|
router |
OpenAI-compatible API, worker registry, WebSocket job hub, billing, Redis cache, Supabase persistence, Solana settlement hooks | Node.js, Hono, TypeScript, Redis, Supabase |
programs |
Anchor programs for worker registration, job escrow, SLA receipts, penalties, fee distribution, staking, and governance | Rust, Anchor, Solana, Token-2022 |
worker-desktop |
Cross-platform desktop worker app for operators contributing local compute | Electron, TypeScript, Python, Ollama |
worker-cli |
Headless worker for GPU servers and terminal-first operators | Node.js, TypeScript, WebSocket, Ollama/vLLM |
Customers / Agents
|
| OpenAI-compatible HTTPS API
v
Gridlock Router <--------------------> Worker Clients
Hono API Desktop / CLI / Browser
Redis KV cache Ollama / vLLM inference
Supabase state TTFT + TPOT reporting
|
| optional on-chain settlement
v
Solana Programs
ProviderRegistry | JobScheduler | SLARegistry | SLAEnforcer | FeeCollector | Governance
- OpenAI-compatible API at
https://api.grid-lock.tech - Latency SLAs for realtime, standard, batch, and confidential inference
- Worker marketplace with desktop, CLI, and browser worker clients
- WebSocket dispatch for live job assignment and heartbeat tracking
- Prefill / Decode roles for disaggregated inference routing
- KV-cache warm path via Redis-backed prefix routing
- LOCK settlement for job fees, penalties, staking, burns, and treasury flows
- SLA receipts with TTFT, TPOT, pass/fail state, and penalty metadata
- TEE-ready confidential tier for privacy-sensitive inference
- Wallet-owned API keys and billing flows in the web console
curl https://api.grid-lock.tech/v1/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer YOUR_GRIDLOCK_API_KEY" \
-d '{
"model": "llama-3.1-8b-instant",
"messages": [{ "role": "user", "content": "Explain Gridlock in one sentence." }],
"gridlock": { "sla": "standard" }
}'Use the desktop app for a guided operator experience:
https://github.com/Gridlockcompute/worker-desktop/releases
Or run the headless CLI:
git clone https://github.com/Gridlockcompute/worker-cli.git
cd worker-cli
npm install
npm run build
node dist/index.js --wallet YOUR_SOLANA_PUBLIC_KEYStart with the repo that matches the part of the network you want to work on:
- Build API routes, billing, worker routing, or persistence in
router - Build Solana settlement and governance logic in
programs - Improve the packaged operator app in
worker-desktop - Improve server-side worker operation in
worker-cli
- Website: grid-lock.tech
- Production API: api.grid-lock.tech
- GitHub: Gridlockcompute
- Worker Desktop Releases: worker-desktop/releases
Decentralized inference infrastructure with measurable latency guarantees.