| Hi HN, I kept finding myself writing scripts or manually running prompts to do things like PR reviews, documentation generation, or issue triage on my GitHub repositories. After doing this enough times, I decided to build a small platform to automate these workflows. I recently open sourced it: https://github.com/Njuelle/Codaholiq Codaholiq lets you run AI-powered workflows triggered by GitHub events. You connect a repository and define automations that run when certain events occur, such as: - pull request opened
- push to a branch
- issue created
- cron schedule
- manual trigger Each automation defines: - a trigger
- an AI provider or model (Claude Code, Codex, Gemini, etc.)
- a prompt template
- optional conditions on the GitHub event payload When the event happens, Codaholiq renders the prompt using variables from the event context and dispatches a GitHub Actions workflow that runs the AI task. The platform also tracks executions with: - real-time log streaming
- token usage and cost tracking
- execution history One design goal was self-hosting, so the entire platform can run locally or on your own infrastructure. The architecture is intentionally simple: - GitHub webhooks trigger automations
- jobs are queued using Redis and BullMQ
- the NestJS backend processes the queue and dispatches workflows to GitHub Actions
- AI models are called through providers
- logs and token usage are streamed back to the UI The full platform can run with Docker using Postgres and Redis. Example automation: An automation could trigger when a pull request is opened.
It sends the PR diff to an AI model asking for a code review, then runs the task through a GitHub Actions workflow. Some possible use cases: - automated PR reviews
- generating documentation from diffs
- code refactoring suggestions
- issue triage
- automatic changelog generation It's still early, but the core pieces are working. One thing I'm unsure about is whether this project makes more sense as: - a self-hosted automation tool
- or a hosted version that simplifies setup and management I'm curious how teams here would prefer to run something like this. Any feedback is very welcome. |
The architectural gap I'd think about next: what happens when the model isn't confident about a recommendation? Right now it looks like the automation posts results automatically. Most tools treat this as binary — auto-post or require a human click. But there's a useful middle ground where the agent can say "I'm confident on A-C, but flagging D for human review" before posting anything.
Handling that gracefully without bolting on a separate notification system (Slack webhook, email, etc.) is the hard part. Curious how you're thinking about confidence thresholds and escalation paths.