Connect Ollama

Ollama is the shortest path for an operator who already runs a local model. The text worker detects the standard loopback API, lists its models, and lets you choose one explicitly.

Install or Start Ollama

Keep your existing Ollama installation. If you are starting fresh, install it from ollama.com/download, then pull and test the model using Ollama’s own tools. AI Power Grid does not install, import, or upload the model.

Detect the Endpoint

Start Ollama and confirm the intended model is installed:

ollama list

The standard endpoint is http://127.0.0.1:11434. The worker probes /api/tags; it does not infer availability from a model directory.

Compatible Models

The wizard lists the model identifiers returned by Ollama. Select one exact identifier to advertise. Detection means the model is installed; the local generation test must also succeed before you connect it. Vision input is advertised only when the worker’s capability check or an explicit advanced operator declaration supports it.

Connect and Configure

Download the verified text worker from aipowergrid.io/run, start it using the exact platform instructions, and open its token-protected local wizard.

Select the detected Ollama backend.
Select the exact model Ollama reports.
Run the local backend test.
Review context, output, schedule, and concurrency limits.
Approve the worker-scoped connection in Console.
Start the worker and wait for Online.

The selected model has its own response, context, concurrency, and schedule limits in this single-backend setup. Use a pause window or stop the worker to drain and withdraw it from new Grid work.

Security Boundary

The worker connects outbound to the Grid and defaults both its backend and management UI to loopback. Model files remain in Ollama. Any optional Ollama credential stays in the protected local worker configuration and is not returned by the dashboard after saving. The worker never needs a wallet private key.

Community workers receive plaintext prompts and outputs. Do not advertise a machine for private workloads until confidential execution is available.

Verify on the Grid

Enter the exact worker name or ID in the public check on the worker page. A process that merely started is not proof that Core accepted its registration.

⚠️

Grid jobs share the Ollama service with your local applications. The worker honors its own schedule and concurrency, but it does not generally detect and yield to every unrelated local request automatically.

Troubleshooting

Not detected: confirm ollama list works and Ollama is listening on 127.0.0.1:11434, then select Re-scan.
Detected but the test fails: run the model once in Ollama so it is loaded, then retry the local generation test.
Online but no jobs arrive: confirm the exact advertised model on /run. Demand is variable and registration does not guarantee traffic.

Choose a Backend vLLM