Settings Reference

adronite-agent.model

The default model every new chat session starts with (within the selected provider). Switching the model inside a chat overrides it for that session only; this default is left untouched.

This is a new-session default: each new chat opens with this model. Use the model picker in the chat header to switch the model for a single session. That choice is saved with the session and doesn't change this default or affect chats you've already started. See Models → Default model.

adronite-agent.customModels

Define custom models for local LLMs, corporate endpoints, or specialized deployments.

Add and edit models in Settings → Models → Add Custom Model. Each model offers:

• Basics: display name, model name, API URL, and an optional per-model API key
• Token limits: Max Tokens and Max Output Tokens
• Capabilities: reasoning, images, temperature, and streaming toggles
• Advanced Settings: sampling parameters (temperature, top-p, top-k, min-p, and repetition / frequency / presence penalties) and a reasoning/thinking format for open models

adronite-agent.globalAutoApprove

Skip ALL tool approval prompts automatically.

adronite-agent.autoApproveTools

Granular auto-approval for specific tools by name.

{
  "adronite-agent.autoApproveTools": {
    "read_file": true,
    "find_files_by_name": true,
    "grep_file_contents": true,
    "list_symbols_in_file": true,
    "get_file_diagnostics": true,
    "get_git_status": true,
    "edit_file": false,
    "run_shell_command": false
  }
}

adronite-agent.autoApproveCommands

Terminal command allow-patterns (token-prefix) that run without an approval dialog.

A run_shell_command invocation that cleanly matches one of these prefixes runs without prompting. Commands containing shell control characters (;, &, |, backticks, $(), redirects) never auto-approve.

{
  "adronite-agent.autoApproveCommands": ["npm test", "git status", "ls"]
}

adronite-agent.denyCommands

Terminal command deny-patterns (token-prefix) that are refused outright, no dialog.

Deny rules are absolute: they beat every auto-approve grant and apply inside compound commands too.

{
  "adronite-agent.denyCommands": ["rm -rf", "git push --force"]
}

adronite-agent.contextEngineEnabled

Umbrella toggle for the Adronite Context Engine: concept-graph tools, the Index settings tab, and the knowledge-graph subsystem.

One switch controls the whole engine: when on, the agent can pull in relevant context automatically, the Context Engine settings tab is active, and the code-navigation tools are enabled.

adronite-agent.mcpLazyLoading

"Skinny Tools": make tool usage more token-efficient.

Shown as Skinny Tools under Tools & Integrations. When on, the agent loads the full details of a tool only when it needs it, which can cut token usage significantly. Leave it on unless you notice the agent struggling to pick the right tool.

adronite-agent.skillsEnabled

Enable the Skills system for isolated Docker execution.

• skillsDirectory: default ~/.adronite/skills
• skillsDockerImage: default adronite/adronite-skills:latest
• dynamicSkillsEnabled: default true; auto-captures repo-specific procedures as skills in .codistry/skills/

adronite-agent.telemetryEnabled

Anonymous usage data for improving the extension.

Respects VS Code's global telemetry setting.