Devlog 1 of the Blocks.AI Agent Maker
Hi ! I'm participating to the Blocks.AI hackathon, this means that I need to make an agent on the Blocks.AI network.
I already started an agent that make agent. This agent should be ready in like 1 week (as I speak it's the 30 of may 2026).
The tech stack that I'm using is more simple than you would think!
I'm currently using :
The current workflow I'm using is
- I receive a request from Blocks.AI
- I spin up a new Docker container
- In this Docker container I install Python (logic), NodeJS (for OpenCode), the Blocks.AI CLI (to publish the finished agent to Blocks)
- I send the request to OpenCode with a detailed prompt explaining everything
- When OpenCode finishes I start a new session of OpenCode and I tell him to check everything
- When the second session of OpenCode corerct any bugs, I install all Python dependencies OpenCode is using
- I call the handler.py to make a last test
- When every test passes, I finally
blocks publish --listing public --billing-mode free to make the agent public
- Then I return the "generated" URL (because the Blocks CLI doesn't give an URL but because the agent URLs are just app.blocks.ai/agents/{agent name}, I guess the URL)
- The generated agent will run for 30 minutes on the Docker container
If any of these steps errors out I tell OpenCode to fix them
So this is the whole workflow, I don't think it's that hard.
And if you are wondering why there is SO many tests, it's mostly because the free model I use on OpenCode (it's Big Pickle, basically GLM-4.7) is DUMB
written the 30/05/2026 (saturday 30 of may 2026)
← go home