Runtime policies as an actual gate rather than prompt instructions is the right model. Most frameworks just bolt governance on as a wrapper and hope the model obeys. What I'd want on top of this: observability into why agents are hitting policy blocks, not just that they were blocked.
Yep, totally agree. And Orloj has this built in. Tracks the entire lifecycle of your tasks through traces in real time so you can audit why everything happened good/bad. During your task you can see how many tokens each call used (input/output), and latency for each model/tool call.
Looks interesting. Quick question - one of the biggest challenges with agentic systems in non-deterministic behaviour. Does this framework do anything to address this? Does it help test and validate agent behaviour?
This is where the governance layer of Orloj fits in. You create policies and attach them to agents/tools which are all governed at runtime. These policies could be token guardrails, tool authority, etc. You can then check all of the traces of a task to have an audit trail for debugging (cli or UI). There are also human in the loop approval features that can be applied to make sure things are working correctly before proceeding on tasks.
Nice design and license, probably some good idea in there somewhere! My main beef with all these projects is the monolith ambition. Every single one of these uncountable projects is trying to solve everything and feels like buying into Kubernetes. Even if open source, thats just too much and too heavy. Postgres, nats, workflow engine what the hell are you building here? Half of this will be obsolete when the next agent architectures come out and then? Solve a few core problems properly and in a modular way dont model agent systems after k8s.
Definitely understand the perspective. I think with all software in general there will be a time where you need to adapt to changing technologies/architectures which is why we open sourced the core so we can have contributors build off it. And k8s is loved/hated but it does solve a lot of problems depending on the circumstance.
It depends on what you're trying to build to be honest. For simple tasks Orloj can be a little overkill but it really starts shining when you are trying to setup large task flows that need many agents/tools/policies. Working with Terraform/Kubernettes for years gave a lot of the inspiration for the gitops side of things which we think fits naturally with how agent systems work.
Yes! We visited The Prague Orloj 2 years ago and it's amazing engineering. That's why we named it after it, for how it's coordinating and orchestrating so many complex mechanisms.
(for anyone wondering it's pronounced Or-Loy)
You should check out the Olomouc orloj [1]. Equally technically interesting as the Prague one, but with the added "benefit" of having been adjusted for political correctness under the Communist regime.
Didn't realize the Czechs had so many...The story about the clockmaker on the Prague one was interesting. The king trying to blind him so he could never make another for anyone else...
https://en.wikipedia.org/wiki/Prague_astronomical_clock
And it comes from mutated Latin word "Horologium".
[1] https://en.wikipedia.org/wiki/Olomouc_astronomical_clock