Connection Issues

"Could not connect to Ollama"

The extension cannot reach the Ollama API at ollama.host.

Checklist:

1. Is Ollama running? Open a terminal and run:

   ollama list

   If this fails, start Ollama:

   • macOS/Linux: ollama serve or open the Ollama app
   • Windows: Open the Ollama app from the Start menu

2. Is the host URL correct? Check ollama.host in settings. Default is http://localhost:11434.

3. For remote instances — is the server reachable from your machine?

   curl http://your-server:11434/api/version

4. If using HTTPS or a reverse proxy: check that certificates are valid and the proxy forwards the Authorization header.

"Connection refused" on a remote instance

Ollama only listens on localhost by default. To expose it on all interfaces, set the environment variable before starting:

OLLAMA_HOST=0.0.0.0 ollama serve

Or on macOS, set it in the Ollama app preferences under Environment Variables.

Auth token not working

The auth token is stored in VS Code's encrypted secret store, not in settings.json. To update it:

1. Open Command Palette → "Ollama: Manage Ollama Auth Token"
2. Enter the new token value

---

Models Not Appearing

Local Models panel is empty

1. Make sure at least one model is installed:

   ollama pull llama3.2:3b

2. Click the Refresh button in the Local Models panel
3. Check the Opilot output channel for error messages

Models disappear after VS Code restart

The extension auto-refreshes on startup. If models are missing, Ollama may not have started yet. Click Refresh once Ollama is running.

Running models not shown

Running models are fetched from GET /api/ps. If none appear, the model may have been unloaded. Use Start Model from the sidebar to warm it up.

---

Ollama in Copilot Chat

@ollama doesn't appear in chat

• Ensure the extension is installed and enabled
• Ensure GitHub Copilot Chat is installed and active
• Restart VS Code

Chat responds with the wrong model

Click the model picker icon (✦) inside the Copilot Chat input, select 🦙 Ollama, then pick the model you want. The @ollama participant also lets you change the model via the model picker that appears when you start a conversation.

"Model not found" in chat

The model must be installed locally. Pull it first:

ollama pull <modelname>

Or use the Library panel → right-click → Pull Model.

Thinking tags appear in responses

Some reasoning-capable models (DeepSeek R1, QwQ, etc.) output <think>...</think> blocks. The extension strips these before displaying the response. If you see raw XML-like tags, check that your model name is being recognized as a thinking model:

• Open the output channel: View → Output → Opilot
• Look for Stripping thinking block log lines

If not present, the model name may not match the thinking model pattern. Report it as an issue on GitHub.

---

Inline Completions

No completions appearing

1. Confirm ollama.completionModel is set to an installed model
2. Confirm ollama.enableInlineCompletions is true
3. Check that inline completions are enabled in VS Code:
   • File → Preferences → Settings → search "editor.inlineSuggest.enabled" → set to true
4. Completions are triggered after a short pause while typing — make sure you paused after entering code

Completions are slow

Use a smaller, faster model for completions. Recommended:

• qwen2.5-coder:1.5b
• deepseek-coder:1.3b
• starcoder2:3b

Larger models (7B+) will have noticeable latency even on capable hardware.

---

Modelfiles

"Build failed"

• Open the Opilot output channel for the raw error from ollama create
• Check your Modelfile syntax: the FROM instruction is required
• Ensure the base model is installed: ollama pull <base-model>
• Model names must be lowercase alphanumeric with hyphens only

Modelfiles panel is empty

• Check the ollama.modelfilesPath setting (default: ~/.ollama/modelfiles)
• If the folder doesn't exist, create it: mkdir -p ~/.ollama/modelfiles
• Click Refresh Modelfiles in the panel header

---

Performance

High CPU/memory usage

The extension polls the Ollama API on a timer. If you have many models and a short refresh interval, try increasing:

"ollama.localModelRefreshInterval": 60

Slow Copilot chat responses

Local model performance depends entirely on hardware. Large models (13B+) require a GPU with sufficient VRAM for acceptable speed. For fast responses:

• Use a quantized model (:q4_K_M)
• Use a smaller model family
• Ensure GPU offloading is active — check ollama ps and confirm layers are on GPU

---

Logs and Diagnostics

Opening the output channel

View → Output → dropdown → Opilot

Or run Command Palette → "Ollama: Dump Performance Snapshot" to see timing data.

Server log streaming

When ollama.streamLogs is true, Ollama's own server logs are appended to the output channel. This can show model loading errors, GPU allocation failures, and API request tracing.

Set ollama.diagnostics.logLevel to "debug" for maximum verbosity.

---

Still Having Problems?

1. Check existing issues: github.com/selfagency/opilot/issues
2. Open the output channel and copy the relevant log lines
3. File a new issue with:
   • OS and VS Code version
   • Ollama version (ollama --version)
   • Extension version
   • Relevant output channel logs