Show HN: Horizons – OSS agent execution engine

Posted by JoshPurtell 23 hours ago

Show HN: Horizons – OSS agent execution engine(github.com)

I'm Josh, founder of Synth. We've been working on coding agent optimization with method like GEPA and MIPRO (the latter of which, I helped to originally develop), agent evaluation via methods like RLMs, and large scale deployment for training and inference. We've also worked on patterns for memory, processing live context, and managing agent actions, combining it all in a single stack called Horizons. With the release of OpenAI's Frontier and the consumer excitement around OpenClaw, we think the timing is right to release a v0.

It integrates with our sdk for evaluation and optimization but also comes batteries-included with self-hosted implementations. We think Horizons will make building agent-based products a lot easier and help builders focus on their proprietary data, context, and algorithms

Some notes:

- you can configure claude code, codex, opencode to run in the engine. on-demand or on a cron

- we're striving to make it simple to integrate with existing backends via a 2-way event driven interface, but I'm 99.9% sure it'll change as there are a ton of unknown unknowns

- support for mcp, and we are building with authentication (rbac) in mind, although it's a long-journey

- all self-host able via docker

A very simplistic way to think about it - an OSS take on Frontier, or maybe OpenClaw for prod

14 points | 3 comments

skeptrune 22 hours ago|

Fascinating platform. The API surface is much richer than I would have expected. Ooc, at what size do you think teams typically have use for this? I imagine you have to be running quite a few agents at scale before there's a strong usecase.

JoshPurtell 22 hours ago|

Just like codex or opencode provide strong oss implementations of the core agent loop, our ambition (not achieved! hoping this is a solid start) is to provide a solid oss implementation of the context updating loop, memory, basic database + a backend sync layer. And evals + continual learning + gepa optimization.

Just like everyone can write their own agent, yet many opt for codex/claude code sdk/opencode, we think that at some point in our journey, many will also opt for standard implementations of these patterns, for projects big or small.

Realistically, though, the case for a standardized environment grows a lot stronger when you have multi-agent, permissioned actions, and generally just a lot more state than what you can get away with using only opencode + some glue. Insofar as big teams have ambitious products, they might be more likely to try it

Shindi 7 hours ago||

Really cool project, it looks really useful. We’re moving past manual prompt optimization and considering different options for tuning long horizon tasks. We will likely go with Horizons