Agentai

Features

Goal-Based Development

• Describe what you want to build in plain English
• AI automatically creates a step-by-step plan
• Handles file creation, code writing, testing, and command execution

Multiple AI Providers

• Gemini - Google's advanced AI model
• OpenAI - GPT models with strong capabilities
• OpenRouter - Access to many AI models through one service
• Ollama - Run AI models locally on your machine
• Cloudflare AI Gateway - Enterprise AI with analytics and caching

Easy Configuration

• Project Settings: Store settings in your project folder
• Global Settings: Store settings for all projects
• Environment Variables: Override settings when needed
• Switch Easily: Change AI providers and models anytime

Smart Memory

• Project Memory: Remembers your conversation and project state
• Code Understanding: Analyzes your existing code automatically
• Context Awareness: Keeps track of what you're working on

Safe and Secure

• Command Safety: Validates shell commands before running
• Local Processing: Your code stays on your computer
• No External Servers: Built-in file operations

Interactive Interface

• Terminal UI: Clean, modern interface
• Real-time Updates: See progress as it happens
• Activity Logging: Track what AgentAI is doing

Installation

1. Go 1.22+

2. Clone or navigate to the project

   git clone https://github.com/marcuwynu23/agentai.git
   cd agentai
   

3. Build

   go build -o agentai .
   # Windows: agentai.exe
   

4. Configure (e.g., Gemini)

   agentai config set provider gemini --local
   agentai config set api_key YOUR_GEMINI_API_TOKEN --local
   agentai config set model gemini-2.5-flash --local
   

   Or use .env (see .env.example) with GEMINI_API_TOKEN, etc.

Configuration

• Config file: .agentai/config.json
  • Local: <current-directory>/.agentai/config.json (use --local)
  • Global: ~/.agentai/config.json (use --global)
• Keys: provider, api_key, model, base_url
• Resolution: Explicit --config path → local file → global file → environment variables

Commands:

agentai config show --local     # Show repo config
agentai config show --global    # Show user config
agentai config set provider ollama --local
agentai config set model llama3.2 --local
agentai config set base_url http://192.168.1.55:11434 --local  # Optional

Environment (optional):

• GEMINI_API_TOKEN – Gemini API key (used when provider is gemini and no api_key in file)
• OPENAI_API_KEY – OpenAI key (for openai provider)
• GEMINI_MODEL – Model name (e.g., gemini-2.5-flash)
• REQUEST_DELAY, MAX_RETRIES – Rate limiting
• WORKSPACE_PATH, LOGS_PATH – Paths (defaults: cwd, empty)

Supported Providers

Override any with base_url in config.

Cloudflare AI Gateway Setup

For Cloudflare AI Gateway, you need:

1. Account ID - Find it in the Cloudflare dashboard
2. API Token - Create a token with AI Gateway - Read and AI Gateway - Edit permissions
3. Base URL format - https://gateway.ai.cloudflare.com/v1/{account_id}/{gateway_id}/compat

Example configuration:

agentai config set provider cloudflare --local
agentai config set api_key YOUR_CLOUDFLARE_API_TOKEN --local
agentai config set base_url https://gateway.ai.cloudflare.com/v1/123456789/default/compat --local
agentai config set model openai/gpt-4 --local

Usage

Run the interactive TUI chat:

agentai chat

Then type your goal and press enter to start!

With Ollama (default localhost):

agentai config set provider ollama --local
agentai config set model llama3.2 --local
agentai chat

Development Status

• Multi-provider AI: Implemented (Gemini, OpenAI, OpenRouter, Ollama, Cloudflare AI Gateway) via raw HTTP
• Config: Local/global .agentai/config.json and env
• File operations: Create, modify, read in project directory
• Code generation: AI-generated code with cleanup
• Test creation: AI-generated or template test files
• Command execution: Validated, safe command execution
• TUI chat interface: Implemented with live activity logging

How It Works

1. Config: Load provider, api_key, model, base_url (file + env).
2. Project: New run → AI suggests project name and directory; existing run → reuse.
3. Analysis: Scan codebase; summarize for planner.
4. Plan: AI produces a JSON plan (steps with types and dependencies).
5. Execution: For each step, AI reasons then the MCP client runs file/command/test logic.
6. Memory: Results and conversation saved to <project-name>/.memory.json.

Requirements

• Go 1.22+
• API key for Gemini, OpenAI, OpenRouter, or Cloudflare AI Gateway; or local Ollama (no key)

Documentation

Core Documentation

• USAGE.md – Complete usage guide, providers, configuration, and troubleshooting
• CONTRIBUTING.md – Development guidelines and contribution process
• CHANGELOG.md – Version history and release notes
• LICENSE – MIT license information

Development Resources

• docs/architecture.md – Technical architecture and design decisions

Community Resources

• CODE_OF_CONDUCT.md – Community guidelines and code of conduct
• SECURITY.md – Security policies and vulnerability reporting
• RELEASE-NOTES.md – Detailed release information

Issue Templates

• .github/ISSUE_TEMPLATE/bug_report.md – Bug report template
• .github/ISSUE_TEMPLATE/feature_request.md – Feature request template
• .github/PULL_REQUEST_TEMPLATE.md – Pull request template