Operator docsJungle GridSectionDocsGet started
Explore guides for submitting AI workloads and registering GPU nodes on Jungle Grid. Whether you're running inference, training, or fine-tuning, the platform handles hardware placement for you.
User quickstart
Sign in, start with npx, submit a workload by intent, and verify the result in the portal.
Provider quickstart
Start with npx, register a GPU node, let the CLI install the node-agent runtime, and confirm in the portal.
CLI reference
Commands for job submission, node management, auth flows, and environment configuration.
MCP integration
Connect Jungle Grid to Claude Desktop, Cursor, and other AI hosts. Submit and monitor jobs from inside your agent workflow.
Watch the flow first
Start with the CLI demo if you want the operator flow. Start with the Claude demo if you are wiring agents into workload execution. Both show the same abstraction: describe the workload, then let Jungle Grid handle fit and placement.
See a real AI workload run from an Activepieces workflow through Jungle Grid Ready Paths — from trigger to completed result in 31 seconds, without manually provisioning or selecting GPUs.
A straight operator flow in the terminal: describe the workload, submit by intent, and let Jungle Grid place the run on compatible GPU capacity.
A Claude-driven workflow where a natural-language workload request becomes a real GPU-backed execution path through Jungle Grid.
Quickstart
The same CLI and browser entry points serve both paths. Choose the role that matches what you want to do, then follow the steps.
Session flow
Use the browser to choose the user identity, then start the CLI with npx to submit your first workload by intent instead of by hardware.
Setup
The browser and CLI work together. Account creation, identity choice, and the portal start in the browser. Submission and node management move into the CLI after that.
Shared entry points
CLI basics
Managed fleet operators
Provider-only checks
Guided path
The steps below are the current product truth. They use the existing sign-in pages, portal route, and CLI commands already supported today.
Start with account creation or sign-in. When prompted for a role, choose Run jobs so your first session opens the user view.
Use npx on the machine where you want to submit and inspect workloads from the terminal. You can add a global install later if you want the shorter jungle binary on your PATH.
npx @jungle-grid/cli@latest loginAfter the first browser-backed login, verify that this machine is attached to the right Jungle Grid account before you submit real work.
npx @jungle-grid/cli@latest whoamiDescribe what you want to run, not what GPU you want. The orchestrator handles classification and placement.
npx @jungle-grid/cli@latest submit --workload inference --model-size 7 --image pytorch/pytorch:2.4.0-cuda12.1-cudnn9-runtime --name chat-inferOmit --command to use the image default entrypoint or CMD. Change optimize_for later when you need cost, speed, or balanced behavior.
List your workloads, inspect a specific job when needed, and keep the portal open for a visual confirmation of the same session.
npx @jungle-grid/cli@latest jobsnpx @jungle-grid/cli@latest status <job-id>User path
Before you start
Verification
Success looks like
Quick links
Jungle Grid chooses placement internally. Your submit flow stays focused on workload type, model size, and optimization preference.
Troubleshooting
These are the current friction points the product already expects: device-flow login, identity choice, empty portal states, and provider readiness.
That is still a valid flow. The CLI prints a login URL and a device code, and the login page can be completed from any browser session.
npx @jungle-grid/cli@latest login --no-browserReturn to the account sign-in flow and choose the identity you actually want for this session. Jungle Grid treats user and provider sessions separately.
The portal reflects the active role. Users need to submit a workload before jobs appear. Providers need to register and start a node before the provider table fills in.
Set JUNGLE_GRID_API before npx @jungle-grid/cli@latest login if your CLI should target a different orchestrator URL than the default.
Confirm your dispatch URL is reachable, the host exposes nvidia-smi unless you are simulating, and you can provide payout details when prompted during registration. If the managed runtime needs to be repaired or preloaded, run an explicit node-agent install before retrying start.
npx @jungle-grid/cli@latest node install-agent --forceKeep the standard node-agent flow for self-hosted or VM/root-capable machines. For RunPod Pods, give the orchestrator RUNPOD_API_KEY plus the RunPod cloud, disk, and timeout settings, then restart the orchestrator. When local capacity is unavailable, Jungle Grid creates the real user workload pod directly on RunPod, keeps the job queued until that workload reaches running, broadens across compatible managed GPU tiers when RunPod is temporarily full, and retries with backoff instead of failing immediately.
docker compose -f infra/docker/docker-compose.yml up -d --build redis orchestratorMove into the product
The fastest clean path: choose the right identity, run jungle login, do one real action, and verify it in /portal before you move on.