Forge is an open-source, lightweight agent framework for building AI agents with pluggable components. It provides a clean, modular architecture that makes it easy to create agents with different LLM providers and execution environments.

• 🔌 Pluggable Architecture: Interface-based design for maximum flexibility
• 🤖 LLM Provider Abstraction: Support for OpenAI-compatible APIs with extensibility for custom providers
• 🛠️ Tool System: Agent loop with tool execution and custom tool registration
• 🧠 Chain-of-Thought: Built-in thinking/reasoning capabilities for transparent agent behavior
• 💾 Memory Management: Conversation history and context management
• 🔄 Event-Driven: Real-time streaming of thinking, tool calls, and messages
• 🔁 Self-Healing Error Recovery: Automatic error recovery with circuit breaker pattern
• 🚀 Execution Plane Abstraction: Run agents in different environments (CLI, API, custom)
• 🤖 Headless Mode: Automated CI/CD workflows with git integration and PR creation
• 📚 Automated Documentation: AI-powered documentation updates on every PR
• 📦 Library-First Design: Import as a Go module in your own applications
• 🧪 Well-Tested: Comprehensive test coverage (196+ tests passing)
• 📖 Well-Documented: Clear, comprehensive documentation

go get github.com/entrhq/forge

package main

import (
    "context"
    "log"
    "os"

    "github.com/entrhq/forge/pkg/agent"
    "github.com/entrhq/forge/pkg/executor/tui"
    "github.com/entrhq/forge/pkg/llm/openai"
    "github.com/entrhq/forge/pkg/tools/coding"
)

func main() {
    // Create LLM provider
    provider, err := openai.NewProvider(
        os.Getenv("OPENAI_API_KEY"),
        openai.WithModel("gpt-4o"),
    )
    if err != nil {
        log.Fatal(err)
    }
    
    // Create coding agent with built-in tools
    ag := agent.NewDefaultAgent(provider,
        agent.WithCustomInstructions("You are an expert coding assistant."),
    )
    
    // Register coding tools
    coding.RegisterTools(ag, ".")
    
    // Launch TUI executor
    executor := tui.NewExecutor(ag)
    if err := executor.Run(context.Background()); err != nil {
        log.Fatal(err)
    }
}

• Syntax Highlighting: Full syntax highlighting for diffs and code blocks
• Diff Viewer: Interactive unified diff viewer with color-coded changes
• Command Palette: Quick access to settings, slash commands, and actions (Ctrl+P)
• Interactive Settings: Live configuration of auto-approval, model parameters, and system behavior
• Slash Commands: Built-in commands for common workflows (/commit, /pr, /clear, /help)
• Real-time Streaming: See agent thinking, tool calls, and responses as they happen

File Operations:

• read_file - Read files with optional line ranges
• write_file - Create or overwrite files with automatic directory creation
• list_files - List and filter files with glob patterns and recursive search
• search_files - Regex search across files with context lines

Code Manipulation:

• apply_diff - Surgical code edits with search/replace operations
• execute_command - Run shell commands with streaming output and timeout control

Agent Control:

• task_completion - Mark tasks complete and present results
• ask_question - Request clarifying information from users
• converse - Engage in natural conversation

• Workspace Guard: Prevent operations outside designated directories
• File Ignore System: Respect .gitignore and custom ignore patterns
• Tool Approval: Review and approve tool executions before running
• Auto-Approval Whitelist: Configure safe operations to run automatically
• Command Timeout: Prevent runaway processes with configurable timeouts

• Token-Based Pruning: Automatic conversation history management
• Tool Call Summarization: Condense tool results to preserve context
• Composable Strategies: Mix and match context management approaches
• Threshold-Based Trimming: Keep conversation within model limits

• Automated Commits: Review and commit changes directly from the TUI
• Pull Request Creation: Generate PRs with AI-written descriptions
• Headless CI/CD: Run Forge in GitHub Actions and other CI/CD pipelines
• Automated PR Documentation: Auto-update docs when PRs are opened
• Change Tracking: Monitor file modifications across agent sessions
• Diff Preview: View changes before committing

• Installation - Setup and configuration
• Quick Start - Build your first agent
• Your First Agent - Detailed tutorial
• Understanding the Agent Loop - Core concepts

• Configure Provider - LLM provider setup
• Create Custom Tools - Extend agent capabilities
• Setup PR Documentation - Automated docs workflow
• Manage Memory - Context and history management
• Handle Errors - Error recovery patterns
• Test Tools - Testing strategies
• Optimize Performance - Performance tuning
• Deploy to Production - Production deployment

• Overview - System architecture
• Agent Loop - Agent execution model
• Tool System - Tool architecture
• Memory System - Memory management

• API Reference - Complete API docs
• Configuration - All config options
• Tool Schema - Tool definition format
• Message Format - Message structure
• Error Handling - Error types and recovery
• Performance - Performance characteristics
• Testing - Testing framework
• Glossary - Technical terms

See ADRs for detailed design decisions including:

• Automated PR Documentation
• Headless Git PR Creation
• Headless Git Integration
• XML Tool Call Format
• Auto-Approval System
• Context Management
• Streaming Commands
• TUI Design

Forge is built with a clean, modular architecture designed for extensibility:

pkg/
├── agent/          # Agent core, loop, prompts, memory
│   ├── context/    # Context management strategies
│   ├── git/        # Git integration
│   ├── memory/     # Conversation history
│   ├── prompts/    # Dynamic prompt assembly
│   ├── slash/      # Slash command system
│   └── tools/      # Tool interface and built-ins
├── executor/       # Execution environments
│   ├── cli/        # Command-line executor
│   └── tui/        # Terminal UI executor
├── llm/            # LLM provider abstraction
│   ├── openai/     # OpenAI implementation
│   ├── parser/     # Response parsing
│   └── tokenizer/  # Token counting
├── tools/          # Tool implementations
│   └── coding/     # File and command tools
├── config/         # Configuration management
├── security/       # Security features
│   └── workspace/  # Workspace isolation
└── types/          # Shared types and events

• Interface-First: Clean abstractions for maximum flexibility
• Event-Driven: Real-time streaming of agent activities
• Composable: Mix and match components as needed
• Testable: Comprehensive test coverage (200+ tests)
• Type-Safe: Strong typing with Go's type system

package main

import (
    "context"
    "log"
    "os"

    "github.com/entrhq/forge/pkg/agent"
    "github.com/entrhq/forge/pkg/executor/cli"
    "github.com/entrhq/forge/pkg/llm/openai"
    "github.com/entrhq/forge/pkg/tools/coding"
)

func main() {
    provider, _ := openai.NewProvider(os.Getenv("OPENAI_API_KEY"))
    
    ag := agent.NewDefaultAgent(provider,
        agent.WithCustomInstructions("You are a helpful coding assistant."),
        agent.WithMaxIterations(50),
    )
    
    // Register all coding tools
    coding.RegisterTools(ag, ".")
    
    executor := cli.NewExecutor(ag)
    executor.Run(context.Background())
}

type WeatherTool struct{}

func (t *WeatherTool) Name() string { return "get_weather" }
func (t *WeatherTool) Description() string {
    return "Get current weather for a location"
}
func (t *WeatherTool) Parameters() map[string]interface{} {
    return map[string]interface{}{
        "location": map[string]interface{}{
            "type":        "string",
            "description": "City name",
            "required":    true,
        },
    }
}
func (t *WeatherTool) Execute(ctx context.Context, args map[string]interface{}) (string, error) {
    location := args["location"].(string)
    // Call weather API...
    return fmt.Sprintf("Weather in %s: Sunny, 72°F", location), nil
}

// Register custom tool
ag.RegisterTool(&WeatherTool{})

See examples/ for complete working examples.

• Go 1.21 or higher
• Make (optional, recommended)
• Git

# Clone repository
git clone https://github.com/entrhq/forge.git
cd forge

# Install development tools
make install-tools

# Run tests
make test

# Run linter
make lint

# Build CLI
make build

• make test - Run all tests with coverage
• make lint - Run linters (golangci-lint)
• make fmt - Format code
• make build - Build Forge CLI
• make install - Install Forge CLI
• make clean - Clean build artifacts
• make all - Run all checks and build

# All tests
make test

# Specific package
go test ./pkg/agent/...

# With coverage
go test -cover ./...

# Verbose
go test -v ./pkg/tools/coding/

We welcome contributions! Please see CONTRIBUTING.md for guidelines.

• 🔌 Additional LLM provider implementations (Anthropic, Google, etc.)
• 🛠️ New tool implementations
• 📝 Documentation improvements
• 🐛 Bug fixes and performance improvements
• ✨ New features and enhancements

By participating, you agree to abide by our Code of Conduct.

For security issues, see our Security Policy.

• Core agent loop with tool execution
• OpenAI provider with streaming support
• Advanced TUI with syntax highlighting
• Complete coding toolkit (read, write, diff, search, execute)
• Context management with multiple strategies
• Auto-approval and settings system
• Git integration (commits and PRs)
• Headless mode for CI/CD automation
• Automated PR documentation workflow
• File ignore system for security
• Slash command system
• Self-healing error recovery
• XML/CDATA tool call format
• Comprehensive test suite (200+ tests)

• Additional LLM providers (Anthropic Claude, Google Gemini)
• Multi-agent coordination and handoffs
• Advanced executor implementations (HTTP API, Slack bot)
• Plugin system for third-party tools
• Enhanced memory with vector storage

• Web-based UI
• Agent marketplace and sharing
• Observability and monitoring
• Advanced debugging tools
• Multi-modal support (vision, audio)

See ROADMAP.md for detailed plans.

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

Built with ❤️ as part of the Entr Agent Platform.

• 📖 Documentation - Complete documentation
• 💻 Examples - Working code examples
• 🐛 Issues - Report bugs or request features
• 💬 Discussions - Ask questions and share ideas
• 📝 Changelog - Version history
• 🤝 Contributing - Contribution guidelines