What is AI Dev Station?

24/7 AI software development station. Converts Linear tickets into GitHub PRs via tag-based two-phase workflow.

Architecture

Node topology

Machine	Role	Hardware	Services
Mac Mini 2015	Orchestrator	Intel i7, 16 GB	Docker: postgres, redis, n8n, telegram-bridge, light-router, grafana, prometheus, iphone-pwa
Mac Studio M1 Max	Worker	64 GB	Ollama (brew), Docker: pipeline, opencode
MacBook Air M1	Client	8 GB	OpenCode CLI, VS Code, git
iPhone	Remote	—	Telegram app + PWA dashboard

LLM fallback chain

Studio Ollama (local, free) → DeepSeek ($0.14/M) → Claude Sonnet 4.5 ($3/M)

6-Agent Pipeline Chain

PM → Architect → Coder → Reviewer → Healer (auto-fix ×3 max) → Release

Usage

Tag-based triggering

1. Create a Linear issue
2. Add tag "plan" → pipeline creates plan PR (draft)
3. Add tag "implement" → pipeline reads plan, generates code, commits to same PR

Run pipeline manually

docker exec devstation-studio-pipeline-1 python pipeline.py \
  --issue "AI-123" --repo "user/repo" \
  --title "Fix login timeout bug" --phase full

Use OpenCode on Air

opencode "fix the race condition in process.py"
opencode "add error handling" --model deepseek

Dashboards

URL	Service
http://{TAILSCALE_IP}:3000	Grafana (admin/{GRAFANA_PASSWORD})
http://{STUDIO_TAILSCALE_IP}:8001/healthz	Pipeline status
http://{TAILSCALE_IP}:8080	iPhone PWA dashboard

Telegram Commands

Command	Action
/start	Show welcome message + main menu
/menu	Show main interactive menu
/status	Check pipeline status, active tasks, pending reviews
/pause	Pause pipeline
/resume	Resume pipeline
/approve	Approve current pending review
/reject	Reject current pending review
/llm studio|cloud|auto	Set LLM mode
/help	List all commands

The bot also has inline keyboards for quick actions — tap the buttons in the chat instead of typing commands.