Hi everyone, I want to deploy CrewAI as an API service so that I can send requests and receive responses. What’s the best way to deploy it as an API service? I’m using FastAPI and I wanted to deploy it on Heroku - but it doesn’t fit within the 1GB RAM in Heroku, and more than that is just unreaso…

I have the same dilemma as you, have you tried render or digitalocean? i think also that the cerebras.ai can help you with the velocity of the api model

No, I haven’t tried Render or digitalocean. I think I’ll drop CrewAI altogether. It brings so much overhead, and issues like this are very hard to deal with.

FYI: DigitalOcean cheapest GPU droplete ~ $3.64/hour! I have an outstanding query to them asking if this is when used, or when powered up! As yet no reponse. I take it that you have looked at the link I put in my post above. In essence these issues will probably be the same for most cloud GPU pro…

Yes, thank you. I checked your comment. But I was wondering about the deployment of the CrewAI as an API service, natively in Python. I didn’t want to rewrite everything in another language. Even now, when I’m quite sure I won’t use CrewAI directly, I’m still going to use Python. Best regards, …

Hi @Dragutin_Oreski , Python is the issue, not CrewAI. Any AI app created with Python is going to have the same issue. I’m not a Python dev, but I believe that there are techniques to reduce the container size, even thenn it’s going to be significantly larger than an equivilent Golang, Rust applica…

Hi @joseguzman I have created my crew but am getting the last output of the task only. Did it get a similar issue

same issues to i only get the last output of the task were you able to fix this

After reading through the documentation, I understood that the crew will by default return the last output of the last task. You can have a task that compiles the results from other tasks or, You can get the full output by using the full_output parameter in the crew: @crew def crew(self) → Crew: …

Ohhh, i will try creating a task that compiles results from other task. i’ve also tried to use the full_option but i still got the last output. please can you share your github repo. probably i’ll find what i’m doing wrong

Deploying CrewAI as an API service

CrewAI Community Support

Dabnis October 11, 2024, 8:01am 2

Hi @Dragutin_Oreski ,
I have just wrote a response to another post: Client session memory - #4 by Dabnis that may help your considerations.

Reagrds
Dabnis

Topic		Replies	Views
Hosting crew ai on some other platform CrewAI Community Support	6	1507	January 7, 2025
Flask and CrewAI CrewAI Community Support	13	1252	November 17, 2024
Documentation link for crewAI deployment in Cloud CrewAI Community Support	1	235	May 25, 2025
Using the CrewAI CLI within an existing FastAPI application General	4	2682	January 5, 2026
Serverless Deployment General	2	404	January 8, 2025

Deploying CrewAI as an API service

Related topics