AI Compute.Powered by Everyone.
Intelligently routed to verified devices and GPUs.
Swarm connects AI workloads with computing power from people and businesses around the world, making AI more accessible while rewarding providers for the work their devices complete.
One task. The right hardware. A verified result.
Customers do not manage individual nodes. Providers do not hunt for workloads. Swarm coordinates the entire path between demand and verified compute.
Submit Task
Submit AI workloads through the stable API or point existing model-provider integrations at Swarm.
Have capable hardware?
Put it to work.
Swarm turns eligible laptops, desktops, servers, and phones into provider nodes. The network determines what each device can safely run, routes matching workloads to it, verifies completed work, and records provider earnings in USD.
Test Your Device. See What It Can Earn.
Run a quick browser pre-check to estimate your device level, eligible task lanes, and potential USD-denominated earnings. The installed node app performs the final hardware, model, runtime, and policy verification.
Provider earnings are accounted and displayed directly in USD, with no reference market price.
- Scanning CPU
- Checking RAM
- Measuring storage
- Detecting GPU / NPU
- Estimating runtime class
- Calculating earning range
- Operating system
- —
- Memory / RAM
- —
- CPU cores
- —
- GPU
- —
- Browser storage signal
- Not reported
Provider earnings are recorded directly in USD based on verified completed work. The estimator requests USD-denominated ranges directly and does not use a reference market price or speculative market value.
Run the test to request a USD-denominated earning range based on your likely node level, eligible task lanes, uptime, and network demand. Final eligibility is determined by the installed node app and backend policy.
Levels 1–5 live ladder, Level 6 roadmap
No model required
3B–4B class local models, including Gemma 4 E2B on supported Android devices
7B–8B class local models
Swarm Level 4 Code Pack
Gemma 3 27B primary · Mistral Small 24B fallback · guarded 30B lanes
Premium roadmap tier for high-stakes enterprise planning, complex multi-document reasoning, and legal/financial/technical review with human oversight.
Live Swarm Network
0 heartbeat active
Registered nodes: —
Node locations are approximate and derived from public-safe registered node locations. Availability includes recent heartbeat activity and, when reported by the backend, Android nodes that are FCM wake-ready while intentionally idle on battery. This public view does not expose provider emails, wallet addresses, hardware IDs, or task data.
Drop Swarm into the stack
you already use.
Submit AI workloads through the stable API or point existing model-provider integrations at Swarm. Realtime streaming, durable conversation memory, Projects, verified web retrieval, routing, and billing stay behind one network surface.
/compat/openai/v1Turn the hardware your company already owns into AI infrastructure.
Connect GPUs, servers, workstations, and capable devices into one private compute environment. Swarm intelligently routes workloads across your Hive first, keeping sensitive work inside your organization and reducing unnecessary external inference spend.
Private First. Public When Needed. One Orchestration Layer Across Both.
Keep sensitive workloads on infrastructure your organization already governs.
Use compute already purchased before paying external inference rates.
Burst approved workloads to Public Swarm when private capacity is constrained.
Centralized AI keeps asking for
more physical infrastructure.
New AI demand usually means more power delivery, more cooling, more buildings, and long infrastructure lead times. Swarm explores another path: aggregate useful capacity from hardware that is already deployed.
Centralized compute growth depends on grid connections, electrical upgrades, and large blocks of power becoming available in the right place.
Large centralized clusters require purpose-built thermal management. Distributed devices already operate inside environments designed for their normal heat envelope.
Adding centralized capacity can require land, buildings, networking, and cooling before the first workload runs. Swarm can add capacity from hardware that already exists.
The 1.3% thesis belongs
after the product.
Swarm does not need every device on earth. Our scale thesis is that a small share of already-deployed consumer and private compute could add up to hyperscale-class capacity without requiring every new unit of demand to begin with a new centralized facility.
Build on the network
or contribute the compute.
Join Swarm for customer access, provider beta updates, technical previews, and network availability.