limit-cli 0.0.40

AI-powered terminal coding assistant with REPL and TUI. Multi-provider LLM support, session persistence, and built-in tools.
Documentation

limit-cli

Crates.io Docs.rs License: MIT

AI-powered terminal coding assistant with REPL and TUI.

An intelligent coding assistant that lives in your terminal. Features multi-provider LLM support, session persistence, and 18 built-in tools for file operations, git, and code analysis.

Part of the Limit ecosystem.

Why This Exists

Developers shouldn't have to leave their terminal to get AI assistance. limit-cli brings the power of LLMs directly to your command line with a beautiful TUI, full tool integration, and seamless session management.

Features

  • Multi-provider LLM: Anthropic Claude, OpenAI GPT, z.ai GLM, and local models
  • Two interfaces: Full TUI (default) or simple REPL mode with --no-tui
  • 18 built-in tools: File I/O, bash execution, git operations, code analysis
  • AST-aware search: Structural code matching with ast-grep (Rust, TypeScript, Python)
  • LSP integration: Go-to-definition, find-references
  • Session persistence: Auto-save and restore conversations
  • Token tracking: SQLite-based usage tracking with cost estimation
  • Web search: Fetch current information via Exa AI
  • Markdown rendering: Rich formatting with syntax highlighting
  • File autocomplete: Type @ to quickly reference files with fuzzy matching

Installation

One-line Install (Recommended)

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

via Cargo

cargo install limit-cli

Requirements: Unix-like OS (Linux, macOS), Rust 1.70+

Quick Start

# Set your API key
export ANTHROPIC_API_KEY="your-key-here"

# Start the TUI
lim

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

REPL Mode

lim --no-tui

Configuration

Create ~/.limit/config.toml:

provider = "anthropic"

[providers.anthropic]
model = "claude-sonnet-4-6-20260217"
max_tokens = 4096

Environment Variables

Variable Provider
ANTHROPIC_API_KEY Anthropic Claude
OPENAI_API_KEY OpenAI
ZAI_API_KEY z.ai

Built-in Tools

File Operations

Tool Description
file_read Read file contents (max 50MB)
file_write Create or overwrite files
file_edit Diff-based file editing

Shell & Git

Tool Description
bash Execute shell commands
git_status Show repository status
git_diff Show changes
git_log Show commit history
git_add Stage files
git_commit Create commit
git_push / git_pull Remote operations
git_clone Clone repository

Code Analysis

Tool Description
grep Regex search in files
ast_grep AST-aware code search (Rust, TypeScript, Python)
lsp Language server operations

Web

Tool Description
web_search Search the web via Exa AI
web_fetch Fetch URL content as markdown
browser Browser automation for testing and scraping

Session Management

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

# In the TUI/REPL
/session list      # List all saved sessions
/session new       # Create a new session
/session load <id> # Load a specific session
/share             # Copy session to clipboard
/share md          # Export as markdown file
/share json        # Export as JSON file

File Autocomplete

Type @ in the TUI to quickly reference files:

Read @Cargo.toml and analyze the dependencies

Features:

  • Fuzzy matching powered by Frizbee
  • Respects .gitignore and .ignore
  • Keyboard navigation (↑/↓/Enter/Tab/Esc)

Commands

Command Description
/exit Save session and exit
/clear Clear the screen
/help Show available commands
/model Show current model configuration
/tldr Enable code analysis for this project
/session list List all saved sessions
/session new Create a new session
/session load <id> Load a specific session
/share Copy session to clipboard
/share md Export session as markdown
/share json Export session as JSON

Code Analysis (TLDR)

TLDR provides token-efficient code analysis with 95% savings vs reading raw code. It's opt-in per project for performance reasons.

Enabling TLDR

# In the TUI, run:
/tldr

# Or use the alias:
/warm

This enables code analysis for the current project. The setting persists in ~/.limit/tracking.db.

Permissive Mode

If the AI tries to use code analysis but TLDR isn't enabled, it will ask for permission:

AI: Code analysis (TLDR) is not enabled for this project. 
    Would you like to enable it? Run /tldr to activate.

Available Analysis Types

Type Description
search Find functions/structs by keyword
context Get dependencies and callers
source Get function code (instead of file_read)
impact Find who calls a function
cfg Control flow graph
dfg Data flow graph
dead_code Find unreachable functions
architecture Detect module layers

Example Usage

Ask the AI natural language questions:

  • "What does the process_message function do?"
  • "Who calls handle_request?"
  • "Find all functions related to authentication"
  • "Show me the architecture of this codebase"

Token Tracking

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

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

License

MIT © Mário Idival