Bonsai

🌳 Bonsai is a CLI tool for managing LLM conversation trees. AI's in the name.

Bonsai provides a structured way to explore multiple directions of a conversation, experiment freely, and return to earlier points to continue down different paths without friction. Think of it as Git for your LLM sessions: a system that lets you branch, reseed, and graft ideas without losing track of the bigger picture.

With Bonsai, you can:

• Maintain clean histories
• Compare alternate paths
• Keep your conversations organized as they grow
• Use the visualization tool to better understand the tree

The result is a workflow where creativity and control coexist, making it easy to cultivate ideas, revisit roots, and guide your conversations toward meaningful outcomes.

Installation

Option 1: Install directly (recommended)

go install github.com/aarose/bonsai@latest

Option 2: Build from source

git clone https://github.com/aarose/bonsai.git
cd bonsai
go build -o bai

To make the binary globally available (so you can run bai instead of ./bai):

# Use the Makefile to install to your Go bin directory
make install

# Add Go's bin directory to your PATH if it's not already there
echo 'export PATH=$PATH:$HOME/go/bin' >> ~/.zshrc  # or ~/.bashrc
source ~/.zshrc  # or source ~/.bashrc

# Alternative: manually move to a system directory
sudo mv bai /usr/local/bin/

# Or add the current directory to your PATH (temporary)
export PATH=$PATH:$(pwd)

Note: After running make install, you may need to add $HOME/go/bin to your PATH as shown above.

LLM API Setup

To chat with LLMs, you'll need to set up API keys for your preferred providers:

OpenAI (GPT models)

export OPENAI_API_KEY="sk-your-openai-api-key-here"

Anthropic (Claude models)

export ANTHROPIC_API_KEY="sk-ant-your-anthropic-api-key-here"

Supported Models

• OpenAI: gpt-3.5-turbo, gpt-4, gpt-4-turbo, gpt-4o
• Anthropic: claude-3-haiku, claude-3-sonnet, claude-3-opus, claude-3-5-sonnet

Don't have an API key? See instructions below to generate a fake conversation tree to test bai out.

Usage

Basic Commands

bai              # Run the tool
bai version      # Check version
bai --help       # See help

Conversation Management

# Plant a seed and start a new conversation
bai seed "I want to plan a trip to Japan"

# Add to current conversation (automatically gets LLM response if model is set)
bai "What's the best time of year to visit?" --llm gpt-4

# View all conversation branches
bai seeds

# Show conversation history on this branch
bai log

# Switch to different conversation branch
bai checkout <node-id>

# View branching options
bai offshoots

# Copy some context from one conversation to another
bai cherry-pick <node-id>

Example output:

$bai seeds
🌱 Found 2 seed(s) in the Bonsai garden:

ID: f47ac10b-58cc-4372-a567-0e02b2c3d479
🧠 Model: claude-3-haiku
💬 Message: I want to plan a trip by train from Ottawa to a nearby city. What are my options?

ID: fbb2e9dd-4b0b-4462-93eb-1554db2e9fe1 (contains current working node)
🧠 Model: claude-3-haiku
💬 Message: Hi there, please tell me a knock knock joke

LLM Integration

When you use the --llm flag or set a model on a seed conversation, bai will:

1. Create your user message as a node
2. Automatically generate an LLM response
3. Add the response as a child node
4. Continue the conversation tree

# Start conversation with specific model
bai seed "Help me learn Go programming" --llm claude-3-5-sonnet

# Add message and get response
bai "What are goroutines?"

# Switch models mid-conversation
bai "Explain with a code example" --llm gpt-4

Visualization

Launch visualization (opens on http://localhost:8080)

./bai visualize

Custom port

./bai visualize --port 3000

Custom database

./bai visualize --database ./custom.db

Dev Notes

Key Features

• Uses Cobra framework for CLI structure
• Base command "bai" with help and version subcommands
• Modular command structure in the cmd package
• Makefile for easy building and installation
• Ready for go install to make it globally available

Development Commands

go mod download    # Install dependencies
go build -o bai    # Build binary
make build         # Build using Makefile
make install       # Install globally
make test          # Run tests

Release Process

# 1. Update version in cmd/version.go (e.g., v0.1.0 → v0.2.0)

# 2. Commit and tag the release
git add .
git commit -m "Bump version to v0.2.0"
git tag -a v0.2.0 -m "Release v0.2.0"

# 3. Push to GitHub
git push origin main --tags

# 4. Build cross-platform release binaries
make release

# 5. Add the binaries to GitHub Releases as release assets

Test fixture -- generate fake convo

Run with default database (bonsai.db) Default behavior: Uses ~/.bonsai/bonsai.db (same as CLI tool)

./scripts/generate_fake_data.sh

Or specify custom database path

./scripts/generate_fake_data.sh path/to/your/database.db

Or run the Go script directly

go run scripts/generate_fake_data.go [database_path]

How is bai pronounced?

It is pronounced "bye", as in *NSYNC's "Bye Bye Bye".