Skip to content
Ptah Documentation

Ollama

Ollama (local)

Section titled “Ollama (local)”

Ollama lets you run capable open-weight models locally. Ptah talks to Ollama’s HTTP server on your machine, so chat works fully offline and no traffic leaves your network.

What you need

Section titled “What you need”
  • Ollama installed on your machine.
  • At least one model pulled locally.
  • Sufficient VRAM / RAM for the model (a 7B model needs ~8GB, a 70B model needs 48GB+ or quantization).

Pulling a model

Section titled “Pulling a model”
Terminal window
# Small, fast, good for chat
ollama pull llama3.2


# Strong reasoning
ollama pull qwen2.5-coder:32b


# Anthropic-compatible schema (works best with Ptah)
ollama pull gpt-oss

Configuration

Section titled “Configuration”

  1. Make sure Ollama is running. Verify with:

    Terminal window
    curl http://localhost:11434/api/tags

  2. Open Settings → Providers → Ollama in Ptah.

  3. Confirm the server URL. The default is http://localhost:11434.

  4. Click Refresh models — Ptah lists every model ollama list reports.

  5. Pick a default model.

No API key is required for local Ollama.

Ollama Cloud

Section titled “Ollama Cloud”

If you use Ollama Cloud, switch the provider to Ollama Cloud in the same settings pane and paste your Ollama Cloud auth token. The token is stored in encrypted safeStorage.

Verifying it works

Section titled “Verifying it works”
  1. Open the chat and pick any Ollama model.
  2. Send a prompt.
  3. You should see streaming output and a $0 cost (local inference). In the Execution Tree, the model field should show your local model name.

Troubleshooting

Section titled “Troubleshooting”
  • connection refused — Ollama isn’t running. Start it with ollama serve or from the Ollama tray app.
  • Models list is empty — run ollama list in a terminal. If empty, pull a model with ollama pull <model>.
  • model requires more memory — the model is too big for your machine. Try a smaller variant or a quantized tag (:q4_K_M).
  • Tool calls are ignored — the model probably doesn’t support tool use. Switch to a model advertised with tool support (e.g. qwen2.5-coder, gpt-oss).
  • Custom server URL — if you run Ollama on another machine on your LAN, change the URL in Settings to http://<host>:11434.