@drewzemke/synapse NPM

Synapse

Synapse (sy) is a lightweight command-line utility that serves as a quick conduit to your favorite LLM. It allows you to interact with LLMs directly from your terminal, supporting various input methods and configurable profiles.

Features

Stream LLM responses directly to your terminal
Support for piped input
User-defined profiles with custom system prompts
User-defined models with multiple provider support
Conversation history and continuation
Interactive chat mode

Installation

npm install -g @drewzemke/synapse

Usage

NOTE: Synapse supports multiple LLM providers including Anthropic, OpenAI, and OpenRouter. You'll need to set the appropriate API key for your chosen provider in your environment.

# Set the API key for your preferred provider
export ANTHROPIC_API_KEY=<your-anthropic-api-key>
# or
export OPENAI_API_KEY=<your-openai-api-key>
# or
export OPENROUTER_API_KEY=<your-openrouter-api-key>

# Send a simple query to the LLM (responses stream to the terminal)
sy "What is a binary tree?"

# Use a specific user-defined profile for a query
sy -p coding "Explain recursion"

# Use a specific model defined in your config
sy -m gpt4 "Compare GPT-4 with Claude"

# Mix and match profiles with different models
sy -p coding -m gpt4 "Explain promises in JavaScript"

# Continue the previous conversation
sy -e "Can you provide an example?"

# Continue a conversation using a specific model
sy -e -m claude "Can you elaborate on that?"

# Show the last response from the LLM
sy -l

# Start an interactive chat session 
sy --chat

# Pipe content as context for the query
cat file.js | sy "Explain this code"

# Use verbose mode to see configuration and diagnostic information
sy -v "Explain quantum computing"

Continuing Conversations

Synapse automatically saves your conversation history, making it easy to continue discussions with the LLM. When you use the -e or --extend flag, Synapse will load your most recent conversation and send the entire conversation history to the LLM for context.

# Ask an initial question
sy "What are the SOLID principles in software design?"

# Continue the conversation with follow-up questions
sy -e "Can you explain the first one in more detail?"
sy -e "How does that compare to the Open/Closed Principle?"

Interactive Chat Mode

Synapse offers an interactive chat mode that lets you have back-and-forth conversations without repeatedly invoking the command.

# Start a new chat session
sy --chat

# Start a chat with an initial prompt
sy --chat "Let's talk about functional programming"

# Continue your previous conversation in chat mode
sy --chat -e

# Use a specific profile and model in chat
sy --chat -p coding -m gpt4

# Combine with verbose mode to see token usage
sy --chat -v

Your selected configuration (profile, model, etc) will persist throughout the chat session. Type exit or press Ctrl+C to exit the chat.

Configuration

Synapse can be configured using a TOML file. The configuration file is automatically loaded from:

Linux/macOS: ~/.config/synapse/config.toml
Windows: %APPDATA%\synapse\config.toml

The directory will be created automatically on first run if it doesn't exist.

General Settings

# General settings
[general]
# Whether to stream responses by default (default true)
stream = true

# Whether to output code blocks with fancy styling and color (default false)
color = true

# Which profile to use by default (default undefined)
default_profile = 'base'

# Which model to use by default (default undefined)
default_model = 'claude'

Profiles Configuration

Profiles allow you to define different system prompts and temperature settings for different use cases.

[profiles.base]
system_prompt = "You are a helpful AI assistant."
temperature = 0.7

[profiles.coding]
system_prompt = "You are a coding assistant. Provide concise, practical answers with code examples."
temperature = 0.2

[profiles.creative]
system_prompt = "You are a creative writing assistant. Be imaginative and inspiring."
temperature = 0.9

Model Configuration

You can define custom models to use with different providers. The "provider" field must be one of the supported providers ('anthropic', 'openai', 'openrouter', or "bedrock"), and the "model" field should be a valid model string for that provider. You can define any number of model specifications and name them as you please.

[models.claude]
provider = "anthropic"
model = "claude-3-7-sonnet-latest"

[models.gpt4]
provider = "openai"
model = "gpt-4-turbo"

[models.or-claude]
provider = "openrouter"
model = "anthropic/claude-3.5-sonnet"

[models.bedrock]
provider = "bedrock"
model = "anthropic.claude-3-7-sonnet-20250219-v1:0"
aws_region = "us-west-2" # or set AWS_REGION in your shell environment

For model strings, please refer to each provider's documentation:

Anthropic: https://docs.anthropic.com/claude/docs/models-overview
OpenAI: https://platform.openai.com/docs/models
OpenRouter: https://openrouter.ai/docs
Amazon Bedrock: https://docs.aws.amazon.com/bedrock/

Development

# Install dependencies
pnpm install

# Run in development mode
pnpm dev

# Build the project
pnpm build

# Format code with BiomeJS
pnpm lint
pnpm lint:fix

# Run tests
pnpm test
pnpm test:watch