Settings now use the opilot.* namespace. Legacy ollama.* keys are still supported and automatically migrated on activation.

Open settings via:

• File → Preferences → Settings then search "Ollama"
• Or edit settings.json directly

Connection

opilot.host

Type	Default
string	"http://localhost:11434"

The URL of your Ollama server. Supports local and remote instances.

Examples:

"opilot.host": "http://localhost:11434"
"opilot.host": "https://my-ollama-server.example.com"
"opilot.host": "http://192.168.1.50:11434"

For remote instances that require authentication, also set an auth token via the Manage Ollama Auth Token command — it is stored securely in the VS Code secret store, not in settings.json.

Model Parameters

Per-model model settings (webview)

In addition to global settings, Opilot now supports per-model generation controls through Ollama: Open Model Settings.

These overrides are persisted as JSON in the extension global storage directory:

• <globalStorage>/model-settings.json
• Schema: Record<modelId, Partial<ModelOptions>>

Supported per-model fields include:

• temperature
• top_p
• top_k
• num_ctx
• num_predict
• think
• think_budget

Use the webview for these values rather than editing the file directly.

Sidebar Refresh Intervals

opilot.localModelRefreshInterval

Type	Default
number	30

How often (in seconds) to auto-refresh the local models and running models lists. Decrease for faster status updates; increase to reduce API polling.

This interval also drives the status bar heartbeat polling cadence.

opilot.libraryRefreshInterval

Type	Default
number	21600

How often (in seconds) to auto-refresh the Ollama Library and Cloud model catalogs. Default is 6 hours. These catalogs change infrequently, so aggressive polling is unnecessary.

Logging

opilot.streamLogs

Type	Default
boolean	true

When enabled, streams Ollama server log output to the Opilot output channel in real time.

• macOS: tails ~/.ollama/logs/server.log
• Windows: tails %LOCALAPPDATA%\Ollama\server.log
• Linux: streams from journalctl -u ollama

Disable if stream output is noisy or you prefer a quiet channel.

opilot.diagnostics.logLevel

Type	Default	Options
string	"info"	"debug", "info", "warn", "error"

Controls verbosity of extension diagnostic output in the Opilot output channel.

Level	Shows
debug	All messages including internal timing and API calls
info	Normal operation messages (default)
warn	Warnings and errors only
error	Errors only

Use "debug" when troubleshooting connection or provider issues.

Modelfiles

opilot.modelfilesPath

Type	Default
string	"" (empty)

Path to the folder containing your Modelfiles. Leave empty to use the default: ~/.ollama/modelfiles.

"opilot.modelfilesPath": "/Users/yourname/projects/my-modelfiles"

Inline Completions

opilot.completionModel

Type	Default
string	"" (disabled)

The Ollama model to use for inline code completions. Must be a locally installed model. Leave empty to disable completions.

Best results with small, fast code models:

"opilot.completionModel": "qwen2.5-coder:1.5b"
"opilot.completionModel": "deepseek-coder:1.3b"
"opilot.completionModel": "starcoder2:3b"

opilot.enableInlineCompletions

Type	Default
boolean	true

Master toggle for inline code completions. Set to false to temporarily disable without clearing your completionModel.

"opilot.enableInlineCompletions": false

Recommended Configuration

A sensible starting configuration for local development:

{
  "opilot.host": "http://localhost:11434",
  "opilot.localModelRefreshInterval": 30,
  "opilot.libraryRefreshInterval": 21600,
  "opilot.streamLogs": true,
  "opilot.diagnostics.logLevel": "info",
  "opilot.completionModel": "qwen2.5-coder:1.5b",
  "opilot.enableInlineCompletions": true
}