Q-11: Remote UI / Observability

Status: ANSWERED Agent: codex/ext-agent Timestamp UTC: 2026-05-11T02:50:00Z Claim: CLAIMED 2026-05-11T02:47:00Z

Prior Answers Checked

• Q-01 (pi-teams fit) — establishes pi-teams as main orchestration surface; tmux panes are the primary GUI
• Q-06 (container shape) — single Docker container recommended; operational controls needed
• No prior answer exists for Q-11 specifically

Short Answer

Use SSH + tmux as the primary remote surface today. Forgejo web UI as the secondary repo ledger. pi-agent-dashboard as the aspirational browser UI (requires container runtime). pi-gui is the wrong tool — it is a local-only pi CLI wrapper, not a remote monitoring dashboard.

Stack from simplest to most complete:

Evidence

pi-gui (minghinmatthewlam/pi-gui)

• Installed on Mac via Homebrew (v0.1.0-beta.28)
• Electron desktop app that wraps local pi CLI sessions
• Readme confirms: "Opens local workspaces", "depends on upstream pi package for session management"
• Verdict: Cannot monitor remote Pi agents. Local-only tool.

pi-agent-dashboard (BlackBeltTechnology/pi-agent-dashboard)

• Installed on Mac as DMG (v0.5.1, darwin-arm64)
• Three-component architecture: bridge extension (per-pi-process) -> WebSocket server (port 9999) -> React web client (port 8000)
• Bridge extension auto-registers in ~/.pi/agent/settings.json when dashboard server starts
• Already deployed at /mnt/kitchen/ops/pi-agent-dashboard/ (mounted read-only in all containers at /from/ops/dashboard)
• Verdict: Correct tool, but requires (a) dashboard server running, (b) bridge extension loaded in agent pi processes, (c) network reachability between bridge and server. Not currently wired.

Forgejo Web UI

• FROM Forgejo at http://100.120.38.37:3001 (container from-forgejo)
• Port 3001 -> container:3000, SSH on 2223 -> container:2223
• Currently down (all Docker containers crashed under load ~2026-05-11)
• Forgejo binary exists inside Docker image, not natively on host
• Verdict: Best repo browser when running. Restart only when system load is safe.

SSH + tmux

• ssh root@relik-pi4 works reliably (tailnet, key auth)
• Agents run in Docker containers on bridge network from-house_from-house-net
• Workcell path: /mnt/kitchen/from-house/workspace/d3-tui-pi-teams-proto/
• Verdict: Only currently-working access method. Use tmux attach for panes when pi-teams is active.

Current System State (2026-05-11)

• Pi 4 is under severe load — all 7+ Docker containers crashed
• Only a single ephemeral d3-tui-pi-teams-proto container running (4s old at last check)
• SSH still works but Docker-dependent services (Forgejo, agents, Jellyfin) are all down

Fit For This Pi Workcell

• Three Pi agents: Best monitored via tmux panes (pi-teams native surface). pi-agent-dashboard can show live reasoning when wired.
• External agents: Need the workcell prompt + wiki paths, not a UI. Already handled by Q-12.
• D3-TUI repo: Forgejo web UI is the right viewer. Repo is safe at /mnt/kitchen/from-house/repos/d3-tui/ (11 source files, 21 tests, 10+ commits).
• Forgejo issues: Accessible via Forgejo API/UI when container is running. Alternative: read issues from SQLite DB at /mnt/kitchen/from-house/forgejo/data/forgejo.db.
• LLM-wiki: Accessible via SSH file reads. No UI needed — agents read files directly.
• Pi resource constraints: Every additional service (dashboard server, Forgejo) adds RAM/CPU load. The crash event suggests ~7 containers exceeds the Pi 4 capacity.

Risks / Failure Modes

1. Container overload: Running Forgejo + dashboard server + agents in Docker simultaneously exceeds Pi 4 resources leading to mass container death. Mitigation: staged startup, resource limits, or run Forgejo natively.
2. pi-agent-dashboard wiring gap: Bridge extension must be configured in each agent pi settings to connect to dashboard server. Currently not configured (agents use pi-fork settings without dashboard extension).
3. pi-gui confusion: Installing pi-gui creates false expectation. The tool name suggests "Pi GUI" but it is for local pi CLI, not remote Pi agent monitoring.
4. Tailnet dependency: Forgejo UI only accessible from tailnet. No public URL. Fine for MVP.
5. No health-check dashboard: No simple /health endpoint or status page exists. Must SSH to check docker ps.

Decision Needed From Mehdi

1. Should we run Forgejo natively (without Docker) to reduce container count?
2. Do you want pi-agent-dashboard wired now, or deferred until after architecture questions are answered?
3. Should pi-gui be removed from Mac since it is not useful for remote monitoring?

Next Probe

Run docker start from-forgejo when system load permits, verify web UI at http://100.120.38.37:3001/from-admin/d3-tui, then assess Pi resource usage with docker stats --no-stream.

Health Check Quick Reference

# Is SSH working?
ssh -o ConnectTimeout=10 root@relik-pi4 echo ok

# What containers are running?
ssh root@relik-pi4 "docker ps --format 'table {{.Names}}\t{{.Status}}'"

# Is Forgejo up?
curl -s -o /dev/null -w "%{http_code}" http://100.120.38.37:3001/

# Read latest agent state
ssh root@relik-pi4 "ls -lt /mnt/kitchen/from-house/state/agents/*/pi/ | head -10"

# Check claim board
ssh root@relik-pi4 "cat /mnt/kitchen/from-house/workspace/d3-tui-pi-teams-proto/llm-wiki/wiki/tasks/claim-board.md"