Feather

Feather - A lightweight agent framework


I don't need layers of abstraction. Base LLMs are very capable. Feather lightly defines agents with tools that auto-execute.

Chaining agents together with Feather looks like this:

const timelineData = twitterAgent.run("Get 50 tweets from my AI list and summarize it for me")
const videoScript = videoScriptAgent.run('Create me a video script based on todays AI news:' + timelineData.output)

To install Feather, run the following command:

npm install feather-ai

DEBUG GUI

Feather comes with an optional GUI that displays detailed log of the agent's system prompt, message history, and LLM requests/responses that went into each assistant message.

OPENROUTER

We use OpenRouter for LLM calls, which uses the OpenAI SDK 1:1. This allows us to use ANY model with ease! While it is a centralized service, it is the easiest, most cost effective way to get access to the latest models instantly and switch models with ease. Also, they accept crypto. If OpenRouter ever goes down, we can switch without trouble as the base SDK is OpenAI.

https://openrouter.ai/

OPENPIPE

We use OpenPipe for collecting training data of agents. This is optional, but HIGHLY recommended for any agent that is used in production. Your data is GOLD, make sure to mine it!

https://openpipe.ai/

.env keys required:

## OPENROUTER KEY FOR AI FUNCTIONS
OPENROUTER_API_KEY=

## OPENPIPE KEY FOR DATA COLLE