GetPullRequest (Sia)

Sia is your AI dev assistant that takes tasks from Slack or Discord, handles the grunt work, writes the code, and delivers ready pull requests all while you are offline.

Comment

GetPullRequest (Sia)

Inspiration

Developers spend too much time on small, repetitive coding tasks - bug fixes, refactors, simple features. We wanted to build an AI assistant that handles these while the dev is sleep, so devs can wake up to ready-to-review PRs instead of a growing backlog.

What it does

Sia is an AI-powered coding assistant that:

  • Takes well-defined coding tasks from a web UI, CLI, or chat apps (Slack/Discord)
  • Spins up cloud dev machines with AI agents (Containing Rovo Dev or Claude Code or Kiro or Cursor)
  • Clones your repo, creates a branch, writes the code
  • Runs build/test/lint verification
  • Creates a Pull Request automatically

You queue up tasks, go to bed, and wake up to PRs ready for review.

How we built it

  • Monorepo with Nx for managing multiple apps
  • Fastify backend with REST APIs and WebSocket for real-time log streaming
  • gRPC for bidirectional communication between backend and AI agents
  • Temporal for durable workflow orchestration (job queues, retries, heartbeats)
  • Next.js + shadcn/ui frontend with TanStack Query
  • Claude Sonnet 4.5 as the AI coding engine in Kiro IDE.
  • PostgreSQL + Drizzle ORM for data persistence
  • PropelAuth for authentication

Challenges we ran into

  • Orchestrating long-running AI coding tasks reliably - solved with Temporal workflows
  • Handling orphan jobs when agents disconnect - built heartbeat monitoring and orphan detection
  • Real-time log streaming at scale - WebSocket manager with proper subscription cleanup
  • Keeping the monorepo organized - Nx + strict TypeScript + steering files for guidelines

Accomplishments that we're proud of

  • End-to-end job execution: from task submission to PR creation
  • Resilient job queue with automatic retry and orphan recovery
  • Platform-agnostic chat integration (Slack working, Discord/Teams in review)
  • Clean architecture with specs documenting every component

What we learned

  • Temporal is amazing for durable workflows - worth the learning curve
  • gRPC streaming is perfect for agent communication
  • Spec-driven development (with Kiro) keeps complex projects organized
  • AI coding assistants need guardrails - verification pipelines are essential

What's next for GetPullRequest (Sia)

  • Discord and Microsoft Teams integrations
  • Multi-agent parallel execution
  • Smarter task decomposition (break big tasks into smaller ones)
  • Self-healing: auto-fix when tests fail
  • Public beta launch

Built With

Share this project:

Updates