Launched this week
CanyonOS
The One Platform for Scaling your Agentic Apps
76 followers
The One Platform for Scaling your Agentic Apps
76 followers
CanyonOS is a workflow-aware platform to build, scale, observe and manage agentic applications. It carries every call between your agents, so it can automatically find and scale just the ones holding up your workflow and show how much each agent adds to your costs, with no tracing code to write. Your agents stay plain Python. Open source under AGPL-3.0.
Launch Team
CotypistType up to 50% less in every Mac app. Free to try, and it runs fully on your Mac.
Promoted
Hey Product Hunt! I’m Ravi from the CanyonOS team.
Scaling your multi-agentic workflows has never been this easy! CanyonOS analyzes your workflow structure and deploys every agent in a separate container in a single-click.
Since CanyonOS carries every call between your agents, it can find and scale the one holding up your workflow, within limits you set. It also enables full observability, showing where requests spent their time and how much each agent adds to your token costs, with no tracing code to write.
It’s built for Python developers deploying and operating multi-agent applications. Your agents stay plain Python, organized to run on CanyonOS, with no SDK or decorators.
This is CanyonOS 0.2, our first public release. It’s free and open source under AGPL-3.0, and you can start locally with Docker or Kubernetes.
Explore the code, try it on your own workflow, and let us know how it goes. Run canyonos feedback to share feedback with the team.
If you're building agent systems in Python, have a look.
@CanyonOS is an o11y platform for managing and scaling multi-agent Python apps at runtime. It carries every call between your agents, so it can find and scale just the ones holding up your workflow and show how much each agent adds to your costs.
It's open source, view source code: github.com/CanyonCodeCoreAI/canyonos
The per-agent cost breakdown is the part I'd actually check first. When I run a few coding agents at once, the slow one is never the one I expect, and the bill is usually where it shows up first. Curious how clean that attribution stays once agents start calling each other in loops — does the cost follow the whole chain, or land on the last call?
@jcbuilds Hey there! the cost is able to follow the whole chain, and different calls to the same agent in the same request would each get logged separately. Every call gets its own metrics about that call, and the rising bill for your agents would sit squarely on the bottlenecked agent.