search

Mintplex-Labs / anything-llm

The all-in-one Desktop & Docker AI application with built-in RAG, AI agents, and more.
https://anythingllm.com
MIT License
24.75k stars 2.5k forks source link

[FEAT] Integrate CrewAI into @agents #1887

Closed jeffscottward closed 3 months ago

jeffscottward commented 3 months ago

Tasks, Roles, Tools, etc etc

timothycarambat commented 3 months ago

CrewAI is a framework for programming agents in python, not a "product", so its not like we can enable you to "port" an agent from crew into AnythingLLM since their only way to invoke crewAI agents via API is via their paid offering :(

We are a NodeJS project, so its not like we can port their code directly into our codebase. If the question is "I want to be able to code an agent like i can with crew" then yes of course that would be coming when we do "custom agents" đź‘Ť

jeffscottward commented 3 months ago

Understood! Thanks for the clarity. :) It’s actually really great because I’m a JavaScript person really myself.

When you do custom agents, I hope that you integrate Mermaid diagrams or have some sort of bridge over Obsidian, it being electron/JS based too.

Also… “something something Fabric prompt library”. Thinking if I should have a Fabric pipeline in obsidian or here.

Jeff Scott Ward Founder / A.I. & U.I. Engineer F0.ai

On Thu, Jul 18, 2024 at 12:17 PM Timothy Carambat @.***> wrote:

CrewAI is a framework for programming agents in python, not a "product", so its not like we can enable you to "port" an agent from crew into AnythingLLM since their only way to invoke crewAI agents via API is via their paid offering https://www.crewai.com/crewaiplus :(

We are a NodeJS project, so its not like we can port their code directly into our codebase. If the question is "I want to be able to code an agent like i can with crew" then yes of course that would be coming when we do "custom agents" đź‘Ť

— Reply to this email directly, view it on GitHub https://github.com/Mintplex-Labs/anything-llm/issues/1887#issuecomment-2237011263, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAJUO7TB3RHTTVQNJQYOBLLZM7TC5AVCNFSM6AAAAABLCEDHYOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDEMZXGAYTCMRWGM . You are receiving this because you authored the thread.Message ID: @.***>

timothycarambat commented 3 months ago

The current plan/concept we have internally is basically a flowise-like node/graph builder. Not flowise itself obviously since that is quote more elaborate, but a UI like that

jeffscottward commented 3 months ago

Very cool!

The main thing I would definitely suggest is having some sore of text syntax for wiring them together.

A UI would just be a nice to have on top IMO. LLM agents that are rapid fire talking to each other would need a syntax to reflect, update, auto gen etc.

Anything where you can have some sort of state machine / diagramming syntax in just text whether it’s Xstate or Mermaid or Diagram-as-Code (eraser.io) or DSL etc would be the building block if I was designing the UX of this (Mermaid my choice, plays well in markdown). Crew has a plan for this already.

But yea excited to see where this all goes. Thank you for doing all this!

Jeff Scott Ward Founder / A.I. & U.I. Engineer F0.ai

On Thu, Jul 18, 2024 at 4:00 PM Timothy Carambat @.***> wrote:

The current plan/concept we have internally is basically a flowise-like node/graph builder. Not flowise itself obviously since that is quote more elaborate, but a UI like that

— Reply to this email directly, view it on GitHub https://github.com/Mintplex-Labs/anything-llm/issues/1887#issuecomment-2237462796, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAJUO7UOTNZF22FFJUHESOLZNANG7AVCNFSM6AAAAABLCEDHYOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDEMZXGQ3DENZZGY . You are receiving this because you authored the thread.Message ID: @.***>