Mermaid - Open Source AI Pair Programmer

An open-source AI pair programmer CLI that provides an interactive chat interface with full agentic coding capabilities. Uses local Ollama models for fast, private, and efficient coding assistance.

Features

• Native Tool Calling: Ollama's tool calling API for structured, reliable actions (v0.2.0+)
• Local Model Support: Use Ollama for fast, private code assistance
• Multiple Local Models: Switch between different Ollama models mid-session without losing context
• Project Aware: Automatically loads and understands your entire project context
• True Agency: Can read, write, execute commands, and manage git
• Privacy First: Run 100% locally with Ollama - your code never leaves your machine
• Interactive TUI: Beautiful terminal interface with Claude Code-inspired aesthetics
• Real-time Streaming: See responses as they're generated
• Smart Context: Respects .gitignore and intelligently manages token limits
• Web Search: Integrated local Searxng for documentation and current information
• Rootless Containers: Secure Podman deployment with no daemon overhead

Quick Start

Prerequisites

• Rust toolchain (required for building from source)

  curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
  source $HOME/.cargo/env
  

• Ollama (required for running local AI models)

  curl -fsSL https://ollama.ai/install.sh | sh
  

• Podman (optional, for web search via Searxng)

  # Ubuntu/Debian/Linux Mint
  sudo apt-get update && sudo apt-get install -y podman podman-compose
  

Installation

Quick Install (Recommended)

If you already have Rust:

cargo install mermaid-cli

If starting from scratch (installs everything):

curl -fsSL https://raw.githubusercontent.com/noahsabaj/mermaid-cli/main/scripts/install.sh | bash

This one-liner installs:

• Rust (if needed)
• Ollama (if needed)
• Mermaid CLI from crates.io
• llama3.1:8b model (4.7GB, tool calling compatible)
• Configures your PATH

After installation, just run:

mermaid

Manual Install (Advanced)

Step-by-step installation with full control:

# 1. Install Rust
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
source $HOME/.cargo/env

# 2. Install Ollama
curl -fsSL https://ollama.ai/install.sh | sh

# 3. Install Mermaid from crates.io
cargo install mermaid-cli

# 4. Download a compatible model
ollama pull llama3.1:8b

# 5. Run Mermaid
mermaid

Updating Mermaid

To update to the latest version:

# Update from crates.io
cargo install mermaid-cli --force

# Or use the one-liner installer (also updates Ollama if needed)
curl -fsSL https://raw.githubusercontent.com/noahsabaj/mermaid-cli/main/scripts/install.sh | bash

Basic Usage

# Start Mermaid with default model
mermaid

# Use a specific model
mermaid --model llama3.1:8b

# List available models
mermaid list

Interactive Commands

Once in the chat interface:

• i - Enter insert mode (type your message)
• Enter - Send message (in insert mode)
• Esc - Return to normal mode
• : - Enter command mode
• Tab - Toggle file sidebar
• Ctrl+C - Quit

Command Mode

• :help - Show all commands
• :model <name> - Switch to a different model
• :clear - Clear chat history
• :sidebar - Toggle file tree
• :quit - Exit Mermaid

Configuration

Environment Variables (.env file)

Set your default model configuration:

MERMAID_DEFAULT_MODEL=ollama/tinyllama

Application Configuration

Located at ~/.config/mermaid/config.toml:

[default_model]
name = "ollama/deepseek-coder:33b"  # provider/model format
temperature = 0.7
max_tokens = 4096

[ui]
theme = "dark"
show_sidebar = true

[context]
max_files = 100
max_context_tokens = 75000

Project Configuration

Create .mermaid/config.toml in your project root to override global settings.

Model Compatibility

Mermaid uses Ollama for local model support with native tool calling (v0.2.0+).

Verified Compatible Models (Tool Calling Support)

Models with native Ollama tool calling support that can execute file operations, commands, and git actions:

Recommended for Coding:

• llama3.1:8b - Fast, excellent tool calling (4.7GB)
• llama3.1:70b - Best quality, slower (40GB)
• qwen2.5-coder:7b - Optimized for code (4.7GB)
• qwen2.5-coder:14b - Excellent coding (9.0GB)
• qwen2.5-coder:32b - Elite coding (19GB)
• mistral-nemo:12b - Balanced performance (7.1GB)

Other Compatible Models:

• llama3.2:1b - Ultra-fast, limited capabilities (1.3GB)
• llama3.2:3b - Fast, decent quality (2.0GB)
• firefunction-v2:70b - Specialized for function calling (40GB)

Models Without Tool Calling

These models can chat but cannot execute actions (coming in v0.2.1 with text fallback):

• deepseek-coder:33b - Excellent for code, no tool support
• codellama - Good for code, no tool support
• tinyllama - Ultra-fast, no tool support
• Most other Ollama models

Installing Models

# Install a compatible model
ollama pull llama3.1:8b

# List installed models
ollama list

# Use with Mermaid
mermaid --model llama3.1:8b

Cloud Models (Ollama Cloud)

Access massive models on datacenter hardware:

• qwen3-coder:480b-cloud - 480B params, elite coding
• kimi-k2-thinking:cloud - 1T params, advanced reasoning
• deepseek-v3.1:671b-cloud - 671B params, largest

Note: Cloud models require an API key from ollama.com/cloud

Example Workflows

Code Generation

You: Create a REST API endpoint for user authentication

Mermaid: I'll create a REST API endpoint for user authentication. Let me set up a basic auth endpoint with JWT tokens.

[Creates files, shows code, explains implementation]

Code Review

You: Review my changes in src/main.rs

Mermaid: I'll review the changes in src/main.rs. Let me check the diff first.

[Analyzes code, suggests improvements, identifies issues]

Debugging

You: The tests are failing, can you help?

Mermaid: I'll help you debug the failing tests. Let me first run them to see the errors.

[Runs tests, analyzes errors, fixes issues]

Refactoring

You: Refactor this function to use async/await

Mermaid: I'll refactor this function to use async/await pattern.

[Shows original code, explains changes, implements refactoring]

Features in Action

Agent Capabilities (Native Tool Calling)

Mermaid uses Ollama's native tool calling API for structured, reliable actions:

Available Tools:

• read_file - Read any file (text, PDF, images with vision models)
• write_file - Create or update files in the project
• delete_file - Delete a file from the project directory
• create_directory - Create a new directory
• execute_command - Execute shell commands and see output
• git_status - Check git working tree status
• git_diff - View changes in files
• git_commit - Create commits with proper messages
• web_search - Search the web via local Searxng

How It Works:

1. Model receives tool definitions as JSON Schema
2. Model calls tools when needed (structured function calls)
3. Mermaid executes the tool and returns results
4. Model continues with the context of results
5. All tool calls are shown in the UI with clear summaries

Project Context

Mermaid automatically:

• Scans your project directory
• Respects .gitignore patterns
• Loads relevant source files
• Understands project structure (Cargo.toml, package.json, etc.)
• Manages token limits intelligently

Development

Building from Source

# Clone the repository
git clone https://github.com/noahsabaj/mermaid-cli.git
cd mermaid

# Build debug version
cargo build

# Run tests
cargo test

# Build optimized release
cargo build --release

Architecture

┌─────────────┐     ┌──────────────┐
│   Mermaid   │────▶│    Ollama    │
│     CLI     │     │  Local Server│
└─────────────┘     └──────────────┘
│                          │
└──────────┬───────────────┘
           ▼
       ┌─────────┐
       │  Local  │
       │ Context │
       └─────────┘

Key Components:

• models/ollama_direct.rs - Direct Ollama connection
• agents/ - File system, command execution, git operations
• context/ - Project analysis and context loading
• tui/ - Terminal user interface with Ratatui
• app/ - Configuration and application state

Privacy First:

• All processing happens locally
• Your code never leaves your machine
• No external API calls or cloud dependencies

Comparison

FAQ

Can I use this with my proprietary code?

Yes! With local models (Ollama), your code never leaves your machine.

Does it work offline?

Yes, with Ollama and local models.

Can I add support for other models?

Mermaid uses Ollama for model support. To use additional models:

1. Pull the model with Ollama: ollama pull model-name
2. Use it with Mermaid: mermaid --model ollama/model-name
3. Check available models at ollama.ai

License

Licensed under either of:

• Apache License, Version 2.0 (LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0)
• MIT license (LICENSE-MIT or http://opensource.org/licenses/MIT)

at your option.

Contribution

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.

Acknowledgments

• Built with Ratatui for the TUI
• Uses Ollama for local model support
• Inspired by Aider, Gemini-CLI, and Claude Code

Community

• GitHub Issues: Report bugs or request features

Note: This project is under active development. Expect breaking changes until v1.0.

Made with love by the open source community