Configuration

Every LocalPilot setting, with defaults.

Ask AI about this page

ChatGPT Claude Perplexity Gemini View as Markdown

The easiest way to configure LocalPilot is its own Settings screen — click the gear icon in the chat toolbar or run LocalPilot: Settings. It has a page for each group below, shows whether a value comes from your User or Workspace settings, lets you reset any value, and includes a provider manager with a Test connection button that lists the models each server exposes.

The first time LocalPilot starts without a reachable model server it opens the Welcome / Setup wizard instead, which detects Ollama, LM Studio, llama.cpp and vLLM, recommends a model and sets an approval preset (Careful, Balanced or Autonomous). Run it again any time with LocalPilot: Welcome / Setup, or disable the automatic prompt with localpilot.showWelcomeOnStartup.

Everything is also available in Settings → Extensions → LocalPilot and in settings.json; the keys are listed below.

LocalPilot Settings inside VS Code — Providers page with live connection status (Ollama reachable, others not running) and one-click presets to add more

Expand a provider to edit it and press Test connection — LocalPilot checks the server, measures latency and lists every model it exposes:

Provider editor with a successful connection test against a running Ollama, listing its models

Agent & approvals page with the Careful / Balanced / Autonomous presets and per-action Ask / Allow / Block controls

Models

Setting	Default	Description
`localpilot.providers`	Ollama, LM Studio, llama.cpp, vLLM on localhost	Model servers. See Models & providers
`localpilot.defaultModel`	`""` (auto)	`provider/model`. Empty picks the best tool-capable coder model that fits this machine
`localpilot.models.disabled`	`[]`	`provider/model` ids hidden from pickers, routing and the status bar (toggle in Settings → Models)
`localpilot.models.autoSetup`	`true`	Offer to install Ollama + a starter model when no local model is found
`localpilot.contextWindow`	`32768`	Requested context (Ollama `num_ctx`); capped by the model’s maximum
`localpilot.maxOutputTokens`	`4096`	Max tokens per response
`localpilot.temperature`	`0.2`	Sampling temperature
`localpilot.thinking`	`false`	Extended thinking for reasoning models

Browser

Setting	Default	Description
`localpilot.browser.chromePath`	`""` (auto)	Chromium-based browser used for headless web-page screenshots (`browser_screenshot`, Screenshot a Web Page…). Auto-detects Chrome, Chromium, Edge, Brave, Arc

Routing

Setting	Default	Description
`localpilot.routing.enabled`	`true`	Pick a model per task: vision model for image turns, smallest FIM model for completions, reasoning model in Plan mode, best available model when the default is offline/disabled
`localpilot.routing.vision`	`""` (auto)	`provider/model` for turns with images when the session model cannot see
`localpilot.routing.reasoning`	`""` (keep session model)	`provider/model` used in Plan mode
`localpilot.routing.fast`	`""` (auto)	`provider/model` for read-only subagents (`explore`, `clarifier`, `researcher`). Empty = smallest tool-capable model that fits; `session` = keep the session model
`localpilot.routing.coding`	`""` (keep session model)	`provider/model` for implementing subagents (`coder`, `tester`, `debugger`, `docs`)

See Automatic routing.

Context

Setting	Default	Description
`localpilot.compactionThreshold`	`0.75`	Fraction of the window at which pruning/summarising starts
`localpilot.maxToolOutputChars`	`24000`	Per-tool-output clamp (head + tail kept)
`localpilot.systemPromptAppend`	`""`	Extra instructions appended to every system prompt
`localpilot.skillsPaths`	`[]`	Additional skill folders

Approvals

Setting	Default
`localpilot.approvals.terminal`	`ask`
`localpilot.approvals.write`	`always`
`localpilot.approvals.delete`	`ask`
`localpilot.approvals.network`	`always`

Values: always, ask, never.

Parallelism

Setting	Default	Description
`localpilot.parallel.maxConcurrency`	`4`	Concurrent tool calls / subagents
`localpilot.parallel.maxSubagentDepth`	`2`	Nesting limit
`localpilot.autoClarify`	`true`	Detect vague requests and hand the agent a workspace snapshot (editor, open files, errors, git) so it states its interpretation before acting

Chat history & memory

Setting	Default	Description
`localpilot.chat.location`	`sidebar`	Where the chat lives: `sidebar` (Primary Side Bar), `secondarySidebar` (right, where Copilot Chat sits), `panel` (bottom, next to the Terminal) or `editor` (a tab in the main area). Change it with LocalPilot: Move Chat To… or by dragging the view
`localpilot.sessions.retention`	`50`	Max chats kept per workspace. The oldest idle chats are removed when a new one starts; pinned and running chats never count. `0` keeps everything
`localpilot.memory.enabled`	`true`	Let the agent save/read durable notes (`.localpilot/memory/` and global storage). Manage notes in Settings → Memory & history

Web search

Setting	Default	Description
`localpilot.webSearch.provider`	`duckduckgo`	`duckduckgo` (no key), `searxng`, `brave`, `tavily`
`localpilot.webSearch.searxngUrl`	`http://localhost:8888`	Self-hosted SearXNG
`localpilot.webSearch.apiKey`	`""`	Brave / Tavily key
`localpilot.webSearch.maxResults`	`6`

Inline completions

Setting	Default	Description
`localpilot.inlineCompletion.enabled`	`false`	Ghost-text completions
`localpilot.inlineCompletion.model`	`""`	`provider/model`; empty uses the smallest FIM-capable model that fits this machine
`localpilot.inlineCompletion.debounceMs`	`350`
`localpilot.inlineCompletion.maxTokens`	`128`

MCP

localpilot.mcpServers — see MCP servers.

Example

{
  "localpilot.defaultModel": "ollama/qwen2.5-coder:32b",
  "localpilot.contextWindow": 65536,
  "localpilot.approvals.terminal": "ask",
  "localpilot.inlineCompletion.enabled": true,
  "localpilot.inlineCompletion.model": "ollama/qwen2.5-coder:1.5b-base",
  "localpilot.webSearch.provider": "searxng",
  "localpilot.systemPromptAppend": "Use pnpm, not npm. Prefer functional React components."
}