Show HN: Build agents via YAML with Prolog validation and 110 built-in tools

[fabceolin]

I'm one of the creators of The Edge Agent (TEA). We built this because we needed a way to deploy agents that was verifiable and robust enough for production/edge cases, moving away from loose scripts.

The architecture aims to solve critical gaps in deterministic orchestration identified by *Prof. Claudionor Coelho Jr. (Stanford alum, ML/DL Faculty at Santa Clara Univ., and Senior Fellow for AI at Majestic Labs)* during our work on the Kiroku project.

*Key Technical Features:*

* *Neurosymbolic Native:* We integrated Prolog to logically validate LLM outputs. This combines neural flexibility with symbolic reasoning to help mitigate hallucinations.

* *YAML + Overlays:* Agents are defined in YAML with overlay support (similar to the Kustomize pattern in Kubernetes), making configs testable and reproducible across environments (Dev/Prod) without code duplication.

* *Hybrid Scripting:*

* *Lua:* Embedded in all binaries (Python, Rust, Wasm) for secure, lightweight logic at the Edge.

* *Python:* Full integration for data science workloads.

* *Batteries Included:* We implemented 110+ tools based on Sarwar Alam’s Agentic Design Patterns. https://github.com/sarwarbeing-ai/Agentic_Design_Patterns

* *Polyglot:* Core written in Rust/Python with Wasm support (runs in browser, Docker, or embedded).

* *Observability:* Native hooks for Comet (Opik) to track execution/cost.

The goal is to provide a solid engineering foundation for agents. I’d love to hear your feedback on the Prolog integration and the YAML-based architecture.

Repo: https://github.com/fabceolin/the_edge_agent

Demo (Wasm): https://fabceolin.github.io/the_edge_agent/wasm-demo

[raphaelcangucu]

Hi Fabricio, can I use this as a Judge?

Let me put the scenario here:

I need a truth resolution mechanism, for example who won some sports match.

I input the sources, news , data, etc and the this agent you handle the judging process.

[fabceolin]

Yes, I wrote an article about this: Truth Resolution Agent: A Multi-Source Judicial Framework for Sports Disputes (Senna 1989 Case Study) using llm as a judge and prolog neurosymbolic as a judge

https://fabceolin.github.io/the_edge_agent/articles/truth-re...

[pisrael]

What is the main difference in results of a pure LLM loop?

[fabceolin]

Clean context for each iteration will make the LLM give your better results. Using LLM loop you will full the context faster degrading the LLM responses. Tea supports create a workflow from dot file https://fabceolin.github.io/the_edge_agent/articles/writing-...

[fabceolin]

Clean context for each iteration will make the LLM give your better results. Using LLM loop you will full the context faster degrading the LLM responses.

[thalesac]

can you elaborate more on the human in the loop? would be nice a more comprehensive example

[fabceolin]

We have checkpoints implemented to save the state in the middle of graph navigation and we can restart from there. It's useful to implement interviews process like https://fabceolin.github.io/the_edge_agent/articles/intellig...

[thalesac]

also I didn't get the name, why edge agent? seems like this is an orchestrator, not edge. seems very useful tho

[fabceolin]

The project started to be a Cyclic State Graph orchestrator, statically defined via YAML, leveraging Neurosymbolic validation (Prolog) to ensure deterministic transitions in edge environments. Langraph also it is, but python and the thread mechanism make not suitable for edge environments.

[johnjr]

Very nice job!

[fabceolin]

Thanks