Async Codex Task Dashboard
Local web UI that dispatches parallel Codex agents against a GitHub repo and tracks their async progress in real time.
Difficulty: weekend | Stack: Python, FastAPI, OpenAI Codex API, WebSockets, HTMX, SQLite
Who this is for
Solo devs and small teams who want visibility into multiple concurrent agentic coding tasks without switching between terminal windows.
Build steps
- Scaffold FastAPI app with SQLite job table (id, prompt, status, output, created_at)
- Wire OpenAI Responses API with background_handle_tool_calls to spawn Codex tasks asynchronously; store task IDs
- Poll task status via OpenAI task endpoint every 5s; push updates over WebSocket to frontend
- HTMX frontend: task submission form, live status cards per task, collapsible diff output
- Add basic cost estimator: track token usage per task, show running total vs. configurable budget cap
Risks
- OpenAI Codex API rate limits will throttle heavy parallel usage — need exponential backoff and queue depth cap
- Async task lifetime can exceed 10 min; WebSocket sessions drop on reconnect requiring state rehydration from SQLite
- Diffing agentic file edits against repo HEAD is non-trivial if agent makes multi-file changes — git apply failures possible
Business Angle
Subscription dashboard that gives solo devs a single pane of glass for dispatching and monitoring parallel Codex agents across repos
Customer: Freelance full-stack dev or indie SaaS builder who runs 3-10 concurrent Codex tasks daily, lives in the terminal, and loses track of which agents finished, failed, or need review
Pricing: saas-mrr — $800 MRR in 4 months (40 users × $20/mo)
Full business breakdown →