⚡ CodeAny

A production-grade AI coding agent CLI built in Go

Inspired by Claude Code's architecture — rebuilt from scratch with more features and better performance

---

Features

Feature	Description
🤖 Multi-provider LLM	Anthropic Claude + OpenAI + any OpenAI-compatible API
🛠️ 14 Built-in Tools	Read/Write/Edit files, Bash, Grep, Glob, WebFetch, WebSearch, Git, Todo, Notebook
🔒 Permission System	3 modes (default/auto/plan) + two-stage safety classifier
💾 Auto Memory	Cross-session memory stored as Markdown files (~/.codeany/memory/)
🔌 MCP Support	JSON-RPC 2.0 over stdio — connect any MCP server
🧠 LSP Integration	Auto-inject compiler diagnostics after each file write
🎭 Agent Teams	Spawn parallel sub-agents with git worktree isolation
📦 Plugin System	Extend with skills, agents, hooks, commands, MCP/LSP servers
⚡ Prompt Caching	3-layer caching (system prompt + tools + last tool_result)
🗜️ Auto Compact	Auto-summarize conversation when context hits 85%
🖥️ Rich TUI	Bubbletea UI with streaming, glamour markdown, status bar
📊 Cost Tracking	Real-time token usage + estimated API cost

Why CodeAny vs Claude Code?

• Pure Go binary — no Node.js runtime, instant startup
• Multi-provider — works with Anthropic, OpenAI, local Ollama, or any OpenAI-compatible API
• Deferred tools — low-frequency tools loaded on demand, reducing token overhead ~40%
• Two-stage permission classifier — pattern matching first, Haiku LLM fallback for ambiguous commands
• Configurable everything — permission mode, context limits, model, memory, LSP all configurable

Installation

One-line Install (macOS / Linux)

curl -fsSL https://raw.githubusercontent.com/thinkany-ai/codeany/main/install.sh | sh

The script auto-detects your OS and architecture, downloads the latest binary from GitHub Releases, and installs it to /usr/local/bin.

Manual Download

Download the pre-built binary for your platform from GitHub Releases:

Platform	File
macOS (Apple Silicon)	codeany_darwin_arm64
macOS (Intel)	codeany_darwin_amd64
Linux (x86_64)	codeany_linux_amd64
Linux (ARM64)	codeany_linux_arm64
Windows (x86_64)	codeany_windows_amd64.exe

# Example: macOS Apple Silicon
curl -fsSL https://github.com/thinkany-ai/codeany/releases/latest/download/codeany_darwin_arm64 -o codeany
chmod +x codeany
sudo mv codeany /usr/local/bin/

From Source

git clone https://github.com/thinkany-ai/codeany.git
cd codeany
go build -o codeany .
sudo mv codeany /usr/local/bin/

Using Go Install

go install github.com/thinkany-ai/codeany@latest

Quick Start

# Set API key
export ANTHROPIC_API_KEY="sk-ant-..."

# Launch interactive mode
codeany

# Start with a prompt
codeany "review the code in this directory and suggest improvements"

# Non-interactive (pipe-friendly)
codeany --print "what does main.go do?"

Configuration

Config File: ~/.codeany/config.yaml

default_model: claude-sonnet-4-5
permission_mode: default     # default | auto | plan
max_iterations: 25
context_window: 200000
compact_threshold: 0.85
memory_enabled: true
lsp_enabled: true

models:
  anthropic:
    api_key: ""              # or set ANTHROPIC_API_KEY env var
  openai:
    api_key: ""              # or set OPENAI_API_KEY env var
    base_url: https://api.openai.com/v1

mcp_servers: []

Environment Variables

ANTHROPIC_API_KEY=sk-ant-...
OPENAI_API_KEY=sk-...

Usage

CLI Flags

codeany [flags] [initial_prompt]

Flags:
  -m, --model string    Model to use (e.g. claude-sonnet-4-5, gpt-4o)
  -d, --dir string      Working directory (default: current dir)
  -p, --print           Non-interactive mode, print response and exit
      --mode string     Permission mode: default | auto | plan
      --no-memory       Disable memory system
      --no-lsp          Disable LSP integration
  -v, --version         Show version

Keyboard Shortcuts

Key	Action
Enter	Send message
Shift+Enter / Ctrl+J	Insert newline
Ctrl+C / Ctrl+D	Exit
Ctrl+L	Clear screen
Up / Down	Scroll history

Slash Commands

Command	Description
/help	Show help
/clear	Clear conversation
/model <name>	Switch model
/cost	Show token usage & cost
/skills	List available skills
/compact	Manually compact conversation
/plan	Switch to plan (read-only) mode
/auto	Switch to auto mode

Built-in Skills

Trigger with /skillname:

Skill	Description
/commit	Generate conventional commit message from git diff and commit
/pr	Generate PR title + description from diff against main
/review	Code review of staged/unstaged changes
/init	Scan project and generate CODEANY.md config file

Permission Modes

Mode	Behavior
default	Safe tools auto-run; dangerous + write ops require confirmation
auto	All tools auto-run (deny rules still apply); suitable for CI
plan	Read-only sandbox; write + dangerous ops silently blocked

Tools

Tool	Type	Description
read	safe	Read file contents with line numbers, offset/limit pagination
write	write	Write/create files (auto mkdir)
edit	write	Precise string replacement (errors if not unique)
bash	dangerous	Execute shell commands (120s timeout, output truncated >500 lines)
grep	safe	Regex search (self-implemented, 3 output modes)
glob	safe	Find files by pattern
list_dir	safe	List directory with git status
web_fetch	safe	Fetch URL and strip HTML
web_search	safe	DuckDuckGo search, top 10 results
git	safe	git status/log/diff/add/commit/push
todo_read	deferred	Read todo list
todo_write	deferred	Write todo list
notebook_read	deferred	Read notebook
tool_search	safe	Discover deferred tools by keyword

Memory System

CodeAny remembers things across sessions using Markdown files in ~/.codeany/memory/:

~/.codeany/memory/
└── {project-hash}/
    ├── MEMORY.md          # Index (auto-injected into system prompt)
    ├── user_prefs.md      # Your preferences and work style
    ├── feedback.md        # Behavior corrections from past sessions
    ├── project_notes.md   # Project conventions
    └── references.md      # Links, boards, docs

Memory files are plain Markdown — you can edit them directly.

MCP Integration

Add MCP servers to ~/.codeany/config.yaml:

mcp_servers:
  - name: filesystem
    command: npx
    args: ["-y", "@modelcontextprotocol/server-filesystem", "/tmp"]
  - name: github
    command: npx
    args: ["-y", "@modelcontextprotocol/server-github"]
    env:
      GITHUB_TOKEN: "${GITHUB_TOKEN}"

MCP tools appear as {server_name}__{tool_name} in the agent.

Architecture

codeany/
├── main.go
├── cmd/            # CLI (cobra)
├── core/           # Agent loop, session, context, prompt builder
├── llm/            # LLM clients (Anthropic, OpenAI, streaming SSE)
├── tools/          # 14 tools + registry + executor pipeline
├── permissions/    # 3-mode permission system + two-stage classifier
├── memory/         # Filesystem-based cross-session memory
├── mcp/            # MCP client (JSON-RPC 2.0 over stdio)
├── lsp/            # LSP client (Content-Length framing)
├── skills/         # Built-in skills + plugin/project skill loading
├── agents/         # Agent Teams + sub-agent execution
├── plugins/        # Plugin manifest loader
├── tui/            # Terminal UI (bubbletea + lipgloss + glamour)
└── config/         # Configuration (viper)

License

MIT — see LICENSE

---

Built with ❤️ using Go, Bubbletea, and the Anthropic API