> For the complete documentation index, see [llms.txt](https://parad0xlabs.gitbook.io/parad0xlabs-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://parad0xlabs.gitbook.io/parad0xlabs-docs/nulla-local-compute-mesh/nulla-local.md). # What It Is ### Your GPU is sitting idle right now. It should be earning. *** ## The Problem With AI Compute Today When you run AI inference (make an AI model generate text, images, or code), you have two options. Both are bad. **Option 1: Pay a cloud AI service.** OpenAI, Anthropic, Google — they own the models and charge a markup. Your API key (your access credential) can be revoked without warning. Your account can be banned. A company you do not control decides whether you can use the tool you are paying for. **Option 2: Rent a GPU from a cloud provider.** AWS, Vast.ai, RunPod — these rent raw computing hardware. You get more control over the software, but the platform still controls job allocation (who gets which machine), takes their cut of every transaction, and monitors your usage. Every path today has a company sitting in the middle, deciding who gets paid and how much. Centralized infrastructure is also fragile: rate-limited (capped on how many requests you can send), monitored, and cuttable. *** ## What NULLA Is **For someone new to crypto:** NULLA is like Airbnb for AI compute — but there is no Airbnb. No company in the middle. Just hosts (people who have hardware) and guests (people who need compute) connecting directly and settling payment automatically. **For crypto builders:** NULLA is a peer-to-peer compute mesh where job routing, credit accounting, and proof-of-work verification run without a central coordinator. Payment settlement uses x402 (an HTTP-native micropayment standard) with USDC (a dollar-pegged stablecoin) on Solana. Proofs are anchored to Solana via SPL Memo instructions (think of SPL Memo as a notary stamp on a public bulletin board — it timestamps your proof on an immutable ledger). Your machine runs a NULLA node. When someone in the mesh needs AI inference, your node sees the job offer, bids on it with a credit price, does the work if selected, and earns NULL credits. No platform picks winners. No oracle (a third-party data feed) decides if you actually did the work — a cryptographic proof does. The proof system uses a challenge-response scheme: the job poster issues a puzzle tied to the task, your node solves it, and the solution is mathematically impossible to fake without actually doing the computation. That proof is then anchored on Solana so anyone can verify it, forever. *** ## What Hardware Works **Apple Silicon** (M1, M2, M3, M4 — the chips in modern MacBooks and Mac Studios), **CUDA GPUs** (Nvidia graphics cards found in gaming PCs and workstations), and standard **CPUs** (the main processor in any computer). The node auto-detects your hardware on startup and only bids on jobs it can handle. A laptop CPU running at roughly 12 tokens per second (tokens are the chunks of text an AI model processes) does not compete for jobs that require an RTX 4090 at 170 tokens per second. The mesh routes intelligently based on advertised capacity and model size — a 7-billion-parameter model and a 70-billion-parameter model are very different workloads, and the router knows the difference. Estimated throughput by hardware tier: | Hardware | Estimated Speed | | ------------------------ | ------------------------------------------- | | Apple Silicon (M-series) | \~55 tokens/second | | Nvidia RTX 3090 | \~100 tokens/second | | Nvidia RTX 4090 | \~170 tokens/second | | Nvidia A100 / H100 | \~200 tokens/second | | CPU-only | \~12 tokens/second (scales with core count) | *** ## What Is Live **Task router — live.** The bid-auction mesh is running. When a job is posted, it broadcasts to all known peers. Peers reply with signed bids. The cheapest valid bid wins. Credits are escrowed (locked up as collateral) before the job starts and released on verified completion. No central server exists anywhere in this path. **Credit ledger — live.** Every earn and spend is recorded in a local SQLite database (a lightweight, file-based database that needs no server) with proof hashes (cryptographic fingerprints) attached to every entry. The ledger is tamper-evident: entries chain together so any modification breaks the chain and is detectable. The ledger can export a proof bundle — a signed record of all your contributions — for future settlement. **Proof-of-work — live.** The two-phase challenge-response system is running. Proofs anchor to Solana via SPL Memo when a keypair (your identity credential on Solana) is configured. Without a keypair, the system runs in dry-run mode and logs the proof hash locally. **Hardware detection — live.** The node probes your hardware on startup using real system tools (psutil for CPU/RAM, nvidia-smi for Nvidia GPUs, platform detection for Apple Silicon) and routes bids accordingly. **In development:** x402 cash settlement (converting NULL credits into USDC you can spend), P2P (peer-to-peer, direct device-to-device) node discovery without a central directory, and rental market broadcasting so your node is visible to the full mesh. *** ## Run a Node Your hardware is already capable. The next section walks through setup — one config file, one command, and you are in the mesh. --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter: ``` GET https://parad0xlabs.gitbook.io/parad0xlabs-docs/nulla-local-compute-mesh/nulla-local.md?ask= ``` The question should be specific, self-contained, and written in natural language. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.