A few months ago this was a sketch. Today we shipped 0.0.1.
The goal is simple. Make it easy for developers to build reliable, planet scale agent workflows without reinventing infra.
Highlights:
• Write agents in clean Python
• Let execution branch at runtime with dynamic DAGs
• Track every step with a durable state manager
• Use built in retries, parallelism, fanout, and joins
• Visualize the execution tree
• Push the same graph from dev to production
If you are exploring agents that plan, adapt, and complete work across huge datasets, I think you will like this release. I would love if you try it. DMs open for feedback and early design partners.
As an example, we built WhatPeopleWant, an open source agent that analyzes Hacker News to surface validated problems for hackers.
Built for devs by devs
Some quick links:
1 - Repo: https://github.com/exospherehost/exospherehost
2 - Docs: https://docs.exosphere.host/
3 - Example Repository: https://github.com/NiveditJain/WhatPeopleWant