readme.md

Flunkydom

The place where flunkies are

This web-application contains all your AI servants (= flunkies) in one place. You can setup several agents which can act independently (like AutoGpt) to achieve goals. To reach these goals, they can use tools and embeddings.

The application is designed to work on multiple goals in parallel and also supports longer-running flows (agents can wait for something, so flows can take days or more to finish. Think of something like Write a tweet about X every 48 hours).

Technicalities

Goals are broken down into tasks (a task = a prompt to GPT and some calculation), which are scheduled asynchronously and will be picked up for execution by a scheduler.

Screenshots

Goals Overview

Embeddings Overview

Details

Agent Templates

You can define custom agents which use an agent template. following templates are available:

General-purpose agent (similar to AutoGPT)

Solves general tasks using chain-of-thought. can use tools to lookup information or execute actions.

Developer Agent (similar to Smol Developer)

writes multi-file programs for a given prompt. It first thinks about files to produce. Secondly, it creates shared dependencies between these files. As a final step, it creates content for each file.

Writing Agent

writes short stories for a given prompt. It first thinks about characters and their relationship. Secondly, it creates a plot and some chapters. As a final step, it creates content for each chapter.

• others planned (for researching topics, for writing etc)

Embeddings

embeddings are created via openai API, stored in postgres using pgvector extension

Tools

• WeatherApi
• Zapier
• serpapi (answerbox/snippet only)
• HomeAssistant (activation of scenes)
• Wait
• ChatGpt

Todo

• abstract underlying ai (so i can switch to something else than openai)

Resources

• https://blog.scottlogic.com/2023/05/04/langchain-mini.html
• https://supabase.com/blog/openai-embeddings-postgres-vector