This is a submission for the MLH x DEV Writing Challenge
What I Built
AI coding agents are great until one gets stuck in a loop and you only notice an hour later. OpenLoom is an open-source, webhook-driven task harness and observer for OpenCode that fixes this.
You send a goal to a simple task API. OpenLoom attaches it to an OpenCode session, watches the session for activity, and streams updates as they happen. It doesn't replace OpenCode. It fills the gaps in OpenCode's HTTP API with:
- a JSON task API
- a session monitor
- webhook notifications
- a web dashboard showing live tasks, per-session token usage, and pending tool approvals (Once / Always / Deny)
If an agent gets stuck, you can abort its session and send a fresh prompt, even from your phone.
My part: I added a new feature to OpenLoom, working solo inside an existing open-source codebase. I built it for the Best Open-Source AI Project challenge in the Hacktoberfest hackathon, submitted through MLH.
Repo: https://github.com/bidyutdolui1970-cpu/my-public-repo
Partner Technologies
OpenLoom talks to OpenCode over HTTP. The stack is Python with a FastAPI backend and a Svelte dashboard. I didn't use any MLH partner tools, so the main technologies were OpenCode and the open-source Python tooling around it.
Getting started takes three commands:
pip install "openloom[ui]"
opencode serve
openloom serve
Then open the dashboard at http://127.0.0.1:55413.
Hackathon Experience
I built this solo during Hacktoberfest 2026, entering the Best Open-Source AI Project challenge through MLH. Working alone meant I owned every step: reading the codebase, designing the feature, and testing it.
The hardest part was debugging. OpenLoom sits between your tools and an AI agent, so when something broke it was hard to tell which side the fault was on. Tracing it step by step taught me to check what the system is actually doing instead of trusting what I assumed it was doing.
What I'll remember most: contributing to a real open-source project is mostly careful reading and patient debugging, and the feature itself is the smaller part.
Top comments (1)
Thanks for reading! 🙌 OpenLoom was my first solo contribution to a real open-source project, and debugging it taught me more than the feature did. If you use OpenCode or other AI coding agents, I'd love to hear how you keep track of them when they get stuck. Feedback and stars on the repo are very welcome!