Configuration

AgentiCode loads configuration from multiple sources with priority: CLI Options > Environment Variables > Config Files > API > Defaults

Config File Locations

Checked in order:

1. .awcode.json (working directory)

2. .awcode/config.json (working directory)

3. ~/.awcode/config.json (home directory)

4. AgenticWork API /api/awcode/config (if connected)

Config File Format

{
  "apiEndpoint": "https://api.agenticwork.io",
  "apiKey": "your-api-key",
  "model": "auto",
  "temperature": 0.7,
  "maxTokens": 4096,
  "maxHistoryLength": 100,
  "maxTurns": 20,
  "features": {
    "shellEnabled": true,
    "fileWriteEnabled": true,
    "webSearchEnabled": true,
    "mcpEnabled": true,
    "codeExecutionEnabled": false
  },
  "mcpServers": [
    {
      "name": "custom-server",
      "command": "node",
      "args": ["/path/to/server.js"],
      "env": {
        "API_KEY": "value"
      }
    }
  ],
  "telemetry": {
    "enabled": true,
    "endpoint": "http://localhost:4318/v1/traces"
  },
  "ui": {
    "theme": "auto",
    "showTokenUsage": true,
    "streamOutput": true
  }
}

Environment Variables

API Configuration

AGENTICWORK_API_ENDPOINT=https://api.agenticwork.io
AGENTICODE_API_KEY=your-api-key
AGENTICWORK_API_KEY=your-api-key  # Alternative
AGENTICWORK_API_URL=https://api.agenticwork.io  # Alternative

Model Configuration

AWCODE_MODEL=gpt-4o
AWCODE_TEMPERATURE=0.7
AWCODE_MAX_TOKENS=4096

Feature Flags

AWCODE_SHELL_ENABLED=true
AWCODE_FILE_WRITE_ENABLED=true
AWCODE_MCP_ENABLED=true

System Prompt

AGENTICODE_SYSTEM_PROMPT="You are a Python expert"
AGENTICODE_SYSTEM_PROMPT_FILE=/path/to/prompt.txt

Session

AGENTICODE_SESSION_ID=session-123

Configuration Options

apiEndpoint

AgenticWork API URL for authentication and features.

"apiEndpoint": "https://api.agenticwork.io"

apiKey

API key for authentication.

"apiKey": "aw_live_xxx..."

model

Default model to use. Can be a model name or preset.

"model": "gpt-4o"
// or
"model": "auto"  // Auto-select based on task

temperature

LLM temperature (0-2). Lower = more deterministic.

"temperature": 0.7

maxTokens

Maximum tokens per response.

"maxTokens": 4096

maxHistoryLength

Maximum messages to keep in history.

"maxHistoryLength": 100

maxTurns

Maximum conversation turns before requiring reset.

"maxTurns": 20

features

Enable/disable specific features:

"features": {
  "shellEnabled": true,       // Shell command execution
  "fileWriteEnabled": true,   // File write operations
  "webSearchEnabled": true,   // Web search tool
  "mcpEnabled": true,         // MCP tool servers
  "codeExecutionEnabled": false // Sandboxed code execution
}

mcpServers

Custom MCP server configurations:

"mcpServers": [
  {
    "name": "my-server",
    "command": "python",
    "args": ["-m", "my_mcp_server"],
    "env": {
      "MY_API_KEY": "secret"
    }
  }
]

ui

UI preferences:

"ui": {
  "theme": "auto",        // "dark", "light", "auto"
  "showTokenUsage": true, // Show token counter
  "streamOutput": true    // Stream responses
}

telemetry

Telemetry configuration:

"telemetry": {
  "enabled": true,
  "endpoint": "http://localhost:4318/v1/traces"
}

Provider-Specific Configuration

Ollama

{
  "provider": "ollama",
  "ollamaHost": "http://localhost:11434",
  "model": "llama3.2"
}

AgenticWork API

{
  "provider": "api",
  "apiEndpoint": "https://api.agenticwork.io",
  "apiKey": "your-key"
}

Auto (Default)

{
  "provider": "auto"
}

Auto-detection priority:

1. AgenticWork API (if configured)

2. Ollama (if running locally)

Project-Level Configuration

Create .awcode.json in your project root for project-specific settings:

{
  "model": "claude-3-5-sonnet",
  "systemPrompt": "You are a TypeScript expert working on this project.",
  "features": {
    "shellEnabled": true,
    "fileWriteEnabled": true
  }
}

AGENTICODE.md

Create AGENTICODE.md in your project root (or run /init) to provide project context:

# Project: My App

## Overview
A React application for...

## Tech Stack
- React 18
- TypeScript
- Tailwind CSS

## Key Files
- src/App.tsx - Main component
- src/api/ - API client code

## Conventions
- Use functional components
- Prefer named exports
- Run tests with `npm test`

This file is automatically loaded and provides context to the AI.