> For the complete documentation index, see [llms.txt](https://docs.usedecentral.org/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.usedecentral.org/providers/native-worker.md). # Native Worker The native worker runs `decentral-node`, a single-binary Rust daemon that manages GPU backends, model downloads, job queue handling, and on-chain settlement interactions. Native workers support larger models, earn more per job, and with an active $DECENTRAL stake receive 85% of each completed job's credit value. *** ## Requirements | Requirement | Details | | ------------- | ------------------------------------------------------------- | | OS | Linux (recommended), macOS, Windows | | GPU | NVIDIA (CUDA), Apple Silicon (Metal), or AMD (ROCm) | | VRAM | 8GB minimum to host the smallest native-tier models | | Solana wallet | Required for registration and receiving USDC payouts | | Disk | 50GB+ free for model storage (70B models require \~40GB) | | Network | Stable connection; lower latency to users improves reputation | *** ## Installation ### Linux (CUDA) ```bash curl -sSL https://install.usedecentral.org | bash ``` The install script downloads the `decentral-node` binary and the CUDA backend. It detects your NVIDIA driver version and selects the correct binary variant. After installation, verify it works: ```bash decentral-node --version # decentral-node 0.1.0 (cuda-12.4) ``` ### macOS (Apple Silicon) ```bash curl -sSL https://install.usedecentral.org/mac | bash ``` The macOS binary uses Metal via llama.cpp. M1, M2, M3, and M4 chips are all supported. Unified memory is used directly, so a 36GB M3 Max can serve 27B parameter models comfortably. ### Windows (CUDA) Download the installer from `usedecentral.org/downloads`. Run the `.exe` installer. The installer adds `decentral-node` to your PATH and installs the CUDA backend. ### AMD (ROCm) ```bash curl -sSL https://install.usedecentral.org/rocm | bash ``` ROCm 5.6+ required. RDNA 2 (RX 6000 series) and later are supported. *** ## Setup ### Step 1: Register your worker ```bash decentral-node register --wallet ``` This command: * Generates a keypair for your worker node (separate from your Solana wallet) * Submits a registration transaction to the `worker_registry` program * Signs the registration using your wallet You will be prompted to approve the registration transaction from your wallet (via QR code or deeplink). The worker keypair is stored at `~/.decentral/worker.json`. Back this up. It is used to sign proof-of-completion submissions and is tied to your on-chain reputation. ### Step 2: Configure models ```bash decentral-node models list # see available models decentral-node models add qwen3-8b decentral-node models add llama-3.3-70b ``` Model weights are downloaded from the Decentral model registry and stored in `~/.decentral/models/`. Downloads are verified against SHA-256 checksums before the model is added to your serving list. To see how much VRAM a model requires before downloading: ```bash decentral-node models info llama-3.3-70b # Model: llama-3.3-70b # Parameters: 70B # Format: GGUF Q4_K_M # VRAM required: 43GB # Tier: Max # Earnings per job: $0.300 (unstaked) / $0.340 (staked) ``` ### Step 3: Stake $DECENTRAL (optional but recommended) Staking is not required to start earning, but it unlocks the 85% payout rate and elevates your routing priority. See the \[Staking guide]\( ) for the full staking flow. ### Step 4: Start the node ```bash decentral-node start ``` The node: 1. Loads your configured models into GPU memory 2. Announces your capabilities to the orchestrator mesh 3. Begins accepting jobs 4. Logs job completions and earnings to the terminal To run as a background service on Linux: ```bash decentral-node install-service systemctl enable decentral-node systemctl start decentral-node ``` *** ## Configuration The configuration file is at `~/.decentral/config.toml`. ```toml [node] wallet = "7xKXtg2CW87d97TXJSDpbD5jBkheTqA83TZRuJosgAsU" worker_keypair = "~/.decentral/worker.json" [models] # Models to load at startup active = ["qwen3-8b", "mistral-7b"] [gpu] # Backend auto-detected from hardware. Override if needed. # Options: cuda, metal, rocm backend = "cuda" # Maximum fraction of VRAM to allocate to decentral-node vram_limit = 0.85 [networking] # Port the worker listens on for orchestrator connections port = 8765 # Optional: set your public IP if auto-detection fails # public_ip = "1.2.3.4" [logging] level = "info" # Log to file in addition to stdout file = "~/.decentral/logs/node.log" ``` *** ## Monitoring The local dashboard is available at `http://localhost:3001` when the node is running. It shows: * Active jobs and queue depth * Earnings per hour and cumulative USDC * GPU utilization and VRAM usage * On-chain reputation score * Per-model job counts and latency metrics You can also query the node's status via its local API: ```bash curl http://localhost:3001/status | jq ``` ```json { "status": "running", "wallet": "7xKXtg2CW87d97TXJSDpbD5jBkheTqA83TZRuJosgAsU", "active_models": ["qwen3-8b", "mistral-7b"], "jobs_completed_24h": 342, "usdc_earned_24h": 27.36, "reputation_score": 847, "stake_active": true, "payout_rate": "85%" } ``` *** ## Updating ```bash decentral-node update ``` The update command downloads the latest binary and replaces the current installation. If you are running as a service, the update command restarts the service automatically after the update. Model weights do not need to be re-downloaded on updates unless a new model format version is released. *** ## Hardware recommendations | GPU | VRAM | Recommended models | Estimated earnings / hour | | -------------------------------- | ----- | ----------------------------------- | ------------------------- | | RTX 3060 | 12GB | Qwen3 8B, Mistral 7B | $0.15 to $0.30 | | RTX 3090 / 4090 | 24GB | Qwen3 8B, Mistral 7B, Llama 3.2 13B | $0.40 to $0.80 | | 2x RTX 3090 (multi-GPU, Q1 2027) | 48GB | Llama 3.3 70B, DeepSeek R1 | $1.50 to $3.00 | | M2 Max 38GPU | 32GB | Qwen3 8B, Llama 3.2 27B | $0.50 to $1.00 | | M3 Ultra 76GPU | 128GB | Llama 3.3 70B, DeepSeek R1 | $1.50 to $3.00 | | A100 80GB | 80GB | All models | $2.00 to $4.00 | Earnings estimates assume sustained job throughput. Actual earnings depend on network demand, model availability, and your reputation score. {: .note } Multi-GPU support (pooling two or more cards per node) is on the Q1 2027 roadmap. The current release uses one GPU per node. --- # 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, and the optional `goal` query parameter: ``` GET https://docs.usedecentral.org/providers/native-worker.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. 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.