Forgecode

AI enabled pair programmer for Claude, GPT, O Series, Grok, Deepseek, Gemini and 300+ models

Last verified:

Visit Forgecode

What is Forgecode?

ForgeCode is the world's top-ranked open-source coding harness, ranked #1 on TermBench 2.0. It is a CLI-based AI coding assistant that allows developers to autonomously write, refactor, debug, and ship code directly from their terminal. Think of it as Claude Code but with first-class support for many AI providers, working equally well with cloud models, open-weight models, and models running locally.

Key features include a multi-agent architecture with bounded context using specialized sub-agents for research, planning, and execution. It offers custom slash commands created with Markdown files in .forge/commands/ to automate repetitive workflows. ForgeCode provides powerful codebase understanding through ForgeCode Services, which includes a context engine for navigating huge codebases, fast tool corrections for local models, and a semantic search engine—all without API keys. The Zsh integration lets you type ':' to talk to ForgeCode right inside your terminal while keeping your custom aliases and Oh My Zsh plugins working.

ForgeCode is designed for developers who want AI pair-programming without leaving their terminal workflow. It supports 100+ LLM providers and models natively from the shell, allowing you to pick the right model for every task—using a thinking model to plan, a fast model to code, and a large-context model for big files. The tool excels at debugging, Git operations, code analysis, and implementing features through natural language prompts.

Forgecode pricing

Pricing model: Freemium

Free tier: $0/forever with basic AI usage and access to default models. Users can use their own API keys (BYOK) from OpenAI, Anthropic, Google, or xAI to unlock premium models. Pro plan: $20/month offering 500 requests/month with access to Claude, OpenAI, Gemini, and other premium models. ForgeCode Services (enhanced codebase understanding, tool-call guardrails, semantic search) is available without API key requirement. Browser-based authentication via app.forgecode.dev with Google or GitHub OAuth is free to set up.

Forgecode pros

  • Ranked #1 coding harness on TermBench 2.0
  • Works with 100+ LLM providers and models natively
  • Multi-agent architecture with specialized sub-agents
  • Custom slash commands via Markdown files for workflow automation
  • Zsh integration with ':' prompt trigger
  • ForgeCode Services provides codebase understanding without API key
  • Supports cloud models, open-weight models, and local models
  • Switch models mid-session without restarting
  • Conversational and intelligent Git operations
  • Semantic search engine for codebases
  • Tool-call guardrails keep local models on track
  • Thousands of evaluations before each change ships
  • Browser-based authentication with Google or GitHub OAuth
  • Project, shared, and global custom command locations
  • Sandbox mode creates isolated git worktrees for experimentation

Forgecode cons

  • Requires Nerd Font installed and enabled in terminal
  • Requires Zsh installed and configured
  • Must restart terminal after Zsh plugin setup
  • Needs at least one AI provider API key or subscription
  • Zsh plugin not active until terminal restart
  • May occasionally introduce unnecessary complexity
  • Learning curve for custom command frontmatter format
  • Windows users need WSL or Git Bash

Frequently asked questions about Forgecode

What is ForgeCode?

ForgeCode is the world's #1 coding harness ranked on TermBench 2.0. It is a CLI-based AI coding assistant that autonomously writes, refactors, debugs, and ships code from your terminal. It works like Claude Code but with first-class support for many AI providers including cloud models, open-weight models, and locally running models.

How do I install ForgeCode?

Install by running 'curl -fsSL https://forgecode.dev/cli | sh' which works on macOS, Linux, Android, and Windows via WSL or Git Bash. Then run 'forge zsh setup' to configure the Zsh plugin, restart your terminal, and run ':login' to authenticate with an AI provider.

What prerequisites do I need?

You need a Nerd Font installed and enabled in your terminal (such as FiraCode Nerd Font) and Zsh installed and configured. The Zsh plugin enables the ':' prompt trigger to talk to ForgeCode directly from your shell.

How do I authenticate with an AI provider?

Run ':login' which walks you through selecting a provider and entering your API key. ForgeCode v0.98.0+ supports browser-based authentication—you run 'forge', it opens your browser to app.forgecode.dev, sign in with Google or GitHub, authorize the app, and authentication completes in under 30 seconds.

What AI providers and models are supported?

ForgeCode supports 100+ providers including OpenRouter (300+ models), OpenAI (GPT Codex series), Anthropic (Claude Sonnet and Opus series), GLM, Kimi, and Minimax. You can use proprietary models like Claude Sonnet/Opus and GPT Codex, or open-source models like GLM, Kimi, and Minimax.

How do custom commands work?

Custom commands are Markdown files in .forge/commands/ directory where the filename becomes the command name. Each file has frontmatter with name and description, plus a body with instructions. Type ':commandname' in chat to execute it. Commands can include prose, bullet lists, XML-style tags for shell commands like <lint> and <test>, and code blocks.

What is ForgeCode Services?

ForgeCode Services provides enhanced codebase understanding with a powerful context engine that navigates huge codebases, fast tool corrections that keep local models on track, and scales over thousands of skills—all without bloating your context window. It includes a semantic search engine and tool-call guardrails, available without API key.

Can I switch models during a session?

Yes, ForgeCode lets you mix and match models without restarting your session. Use ':model' to browse the list, type to filter, and press Enter. You can use a thinking model to plan, a fast model to code, and a large-context model for big files. ForgeCode remembers your choice across sessions.

How does the multi-agent architecture work?

ForgeCode uses specialized sub-agents for research, planning, and execution so each step operates on minimal, relevant context and produces reliable results. The Forge Agent carries out code modifications and executes real-time operations, while the Muse Agent focuses on planning, evaluating, and reviewing code without modifying files.

What CLI commands are available?

Key CLI commands include: 'forge' for interactive REPL, 'forge -p "query"' for direct prompt execution, 'forge --sandbox <name>' for isolated git worktree, 'forge list agents/models/providers/mcp' to list resources, 'forge conversation list/resume/show/info' for session management, 'forge mcp list' for MCP servers, and 'forge --verbose' for detailed logging.

Categories

Use cases

Browse all AI tools on NeedAnAI