Your AI pair programmer that lives in the terminal.

Edit files, run commands, analyze code, manage git — all through natural language. Limit connects to Anthropic Claude, OpenAI, or z.ai to help you code, with a beautiful TUI or simple REPL.

curl -fsSL https://raw.githubusercontent.com/marioidival/limit/trunk/install.sh | bash

cargo install limit-cli

# 1. Clone and build
git clone https://github.com/marioidival/limit.git && cd limit
cargo build --release

# 2. Configure your LLM provider
echo 'provider = "anthropic"' > ~/.limit/config.toml
export ANTHROPIC_API_KEY="your-key-here"

# 3. Run
./target/release/lim

That's it! Start chatting with your AI coding assistant.

Use --no-tui flag for text-based REPL: lim --no-tui

• Multi-Provider LLM Support — Anthropic Claude, OpenAI, z.ai, and local LLMs (Ollama, LM Studio, vLLM)
• 17 Built-in Tools — File I/O, Bash execution, Git operations, code analysis, web search/fetch
• Session Persistence — Auto-save/restore conversation history
• Token Tracking — SQLite-based usage tracking with cost estimation
• Docker Sandbox — Optional containerized tool execution for isolation
• LSP Integration — Go-to-definition, find-references
• AST-Aware Search — Code search that understands syntax (Rust, TypeScript, Python)
• Markdown Rendering — Rich formatting with syntax-highlighted code blocks

Perfect for:

• 🖥️ Developers who live in the terminal
• 🔒 Privacy-conscious teams wanting Docker isolation
• 🏠 Local LLM enthusiasts running models offline
• 📊 Projects requiring audit trails of AI interactions
• 🔄 Multi-model workflows (switch between Claude, GPT-4, z.ai, local)

curl -fsSL https://raw.githubusercontent.com/marioidival/limit/trunk/install.sh | bash

This will install Limit to ~/.local/bin/lim.

git clone https://github.com/marioidival/limit.git
cd limit
cargo build --workspace --release

One-line install:

• curl or wget
• Unix-like OS (Linux, macOS)

From source:

• Rust 1.70+
• git
• Unix-like OS (Linux, macOS)

📚 Quick Setup Guides:

• Anthropic Claude - Recommended for code analysis
• OpenAI GPT - Fast and reliable
• z.ai - Cost-effective alternative
• Local LLMs - Ollama, LM Studio, vLLM
• Full Configuration Guide - All options in one place

Create a configuration file at ~/.limit/config.toml:

provider = "anthropic"

[providers.anthropic]
# api_key is optional - falls back to ANTHROPIC_API_KEY env var
api_key = "sk-ant-api03-..."
model = "claude-sonnet-4-6-20260217"  # or "claude-opus-4-6-20260205" for most capable
max_tokens = 4096
timeout = 60
# Optional: Custom API endpoint
# base_url = "https://api.custom-provider.com/v1/messages"

provider = "openai"

[providers.openai]
# api_key is optional - falls back to OPENAI_API_KEY or ZAI_API_KEY env var
api_key = "sk-..."
model = "gpt-5.4"  # or "gpt-4" for legacy
max_tokens = 4096
timeout = 60
# base_url = "http://localhost:8080/v1/chat/completions"
# Note: Must include full endpoint path (e.g., /v1/chat/completions)

provider = "zai"

[providers.zai]
# api_key is optional - falls back to ZAI_API_KEY env var
api_key = "..."
model = "glm-5"  # 744B parameters, 200K context, best open-weights model
max_tokens = 4096
timeout = 60
# Optional: Enable thinking mode for complex reasoning tasks
# thinking_enabled = false
# clear_thinking = true  # Set to false for Preserved Thinking in multi-turn

provider = "local"  # or "ollama", "lmstudio", "vllm"

[providers.local]
model = "llama3.3"  # or "qwen2.5", "deepseek-r1", "mistral"
base_url = "http://localhost:11434/v1/chat/completions"
# api_key not required for local servers
max_tokens = 4096
timeout = 120  # Longer timeout for local inference

See Local LLM Providers for detailed setup guides.

Provider API keys can be set via environment variables as fallback:

lim

lim --no-tui

If installed from source without the install script: cargo run --package limit-cli

lim> Read the file src/main.rs and explain what it does

[Reading src/main.rs...]

This file contains the main entry point for the CLI application...

See DEVELOPMENT_GUIDE.md for:

• Building and testing
• Project structure
• Adding new tools/providers
• Code style guidelines
• Debugging tips

• Configuration Overview - All configuration options in one place
• Anthropic Claude Setup - Detailed guide for Claude setup
• OpenAI Setup - Detailed guide for GPT-5.4 setup
• z.ai Setup - Detailed guide for z.ai (GLM-5) setup
• Local LLM Providers - Ollama, LM Studio, vLLM, and custom servers

Sessions are automatically saved to ~/.limit/sessions/ and include:

• Conversation history (messages)
• Session metadata (SQLite)
• Agent state
• Token usage statistics

On startup, the last session is automatically restored.

Usage statistics are tracked in ~/.limit/tracking.db:

• Requests count
• Input/output tokens
• Cost estimation
• Duration metrics

Limit can execute tools in isolated Docker containers:

# Requires Docker to be installed and running
# Automatically detected and used if available

Sandbox features:

• 🔒 Isolated execution environment
• 📁 Read-only project mount
• 🚫 No network access
• 💾 512MB memory limit
• ⏱️ 60s timeout

We welcome contributions!

1. Fork and clone the repository
2. Create a feature branch
3. Run tests: cargo test --workspace
4. Run lints: cargo clippy --all-targets && cargo fmt
5. Submit a PR with conventional commit message

MIT