Budibase
AI agents that run your operations
Start new thread
Joe Johnston

Budibase AI Agents - AI agents that run your operations (Open source)

by
Budibase
•
Budibase introduces AI agents for operations teams. Handle requests, approvals, and workflows automatically - connected to your data and tools. Trigger from Slack/Teams/Discord, build apps when needed, and let agents do the work.

Add a comment

Replies

Best
Joe Johnston
Maker
📌
We’ve spent years helping teams build internal tools. But we kept seeing the same thing: Most operations work isn’t about apps. It’s about handling requests. Access requests. Approvals. Repetitive workflows. The stuff that keeps teams busy. So we started asking: What if you didn’t build the workflow at all? What if it just… ran? That’s what led us to AI agents. With Budibase, you can now: • Handle requests automatically (from Slack, forms, etc.) • Run approvals and workflows end-to-end • Connect directly to your data and systems • Build apps only when you actually need them Instead of building tools, you define what should happen. The agent handles the rest. This feels like a new chapter for us. Would love to hear what you think 🙌
Ion Simion Bajinaru

“AI agents that run operations” sounds great.
In reality, most workflows break on edge cases.

Joe Johnston
Maker

Cheers for the comment@ion_simion_bajinaru - it's a fair point and something i've been thinking a lot about recently.

Traditional workflows do break on edge cases because they’re rigid and predefined. Our approach with agents is a bit different - instead of encoding every path upfront, the agent can interpret requests, gather missing context, and adapt when things don’t fit a strict flow.

It’s definitely something we’re still improving (especially in beta), but we’ve found this handles variability much better than static workflows.

Curious, what edge cases have you seen cause the most issues?

Mike Shanks
Maker

@ion_simion_bajinaru 100% agree, and here's what were doing to cover the edge cases:

  • We'll be adding evals asap to help you get the accuracy locked in

  • Our Automations all you to build deterministic workflows. This allows you to allow the agent to what is does best - deal with unstructured data and and fuzzy business rule. The agent can the use your automation as a tool - ensuring 100% determinism in the parts of the process that are already well defined.

Appreciate the comment!

Ion Simion Bajinaru

The real gap in production isn’t interpretation, but control.

Andrew Correa

Is this open source?

Joe Johnston
Maker

@andrew_correa Yes. You can find the repo here:
https://github.com/Budibase/budibase

Taras Shynkarenko

Excited you guys are going to get this 🔥

Mike Shanks
Maker

@tarasshyn Thanks Taras!

Mykola Kondratiuk

Open source AI agents for ops is an interesting wedge - the self-hosted angle matters a lot for anything touching customer data or internal workflows. Most teams I know are cautious about letting a SaaS agent touch their CRM or support queue. How does Budibase handle the agent permissions model - is it role-based on what the agent can read/write, or more of a "here is the whole app, go" setup?

Lien Chueh

Similar to how there could be internal tool sprawl, could there be an issue with AI agent sprawl? For instance, AI agents spun up that do very similar tasks being used within the same team when there should really only be one?

Gyutae Park

Open source + AI agents for ops is an interesting combo. Most no-code tools lock you into their ecosystem hard. How much customization is possible once the agent creates the initial workflow? Can you drop down to code when needed?

Epari Satya Sundar Patra

It's the open-source and self-hosted aspect that really makes this interesting. I've been down the path of writing my own AI agents from scratch with WebSockets and FastAPI and managing the backend logic for running tools and routing data gets really tedious really quickly. Being able to just declare the workflow and have integrations with Slack and database connections out of the box would be incredibly helpful. The thing I'm curious about is whether or not the strictly block based UI hinders the ability to inject some deeply custom logic or components.