A self-hosted command center for AI coding agents. Monitor, orchestrate, and talk back to Claude Code / Codex CLI sessions across every machine you run them on — from a single dashboard.

If you run AI coding agents and you're tired of:

• Alt-tabbing between terminals to see which agent is stuck
• Not knowing what your agent just did while you were in a meeting
• Paying attention to HITL approvals that should happen on your phone
• Re-reading a session transcript to figure out why the build failed

AgentPulse is for you.

• Dashboard — every session (Claude Code + Codex) across every machine, live. Status badges, current-task summary, tool-use counts, plan progress.
• Multi-machine — hooks fire from any laptop / VM / container and stream to one place. Relay on macOS tunnels localhost-only hooks through to a remote server.
• AI watcher — opt-in per-session autonomous loop that reads the transcript, decides what to do next (continue / ask / report / stop), and delivers HITL prompts to your phone.
• Projects — first-class projects with cwd-based session resolution. Templates inherit project defaults with per-field overrides. Auto-created on template save when no project exists for the cwd.
• Ask assistant — chat with your session state. Now also a command surface: launch sessions, edit projects, set up alert rules, archive in bulk, summarize past sessions — all in natural language with HITL approval. "How's the auth refactor going?" / "Any stuck agents?" / "Status across all active tabs?"
• Telegram integration — DM the bot to ask questions or approve HITL requests. Inline buttons, grounded replies.
• Self-hosted — Bun + Hono + SQLite. Single container. Runs on a $100 mini PC.

• Getting Started — install, wire up your first agent, see sessions stream in
• Architecture — event flow, data model, where state lives
• Projects — projects registry, template linkage, alert rules
• AI Watcher — autonomous session observer + HITL loop
• Ask Assistant — the global / Telegram chat on top of your session state
• Telegram Channel — bot setup, webhook vs. polling, Ask-via-Telegram
• Deployment — Docker, Kubernetes, reverse-proxy auth
• Contributing — repo layout, dev setup, conventions, where to start
• FAQ · Roadmap

AgentPulse is built for home-labbers running AI tooling for themselves and developers building agentic infrastructure. It's dogfooded daily on a Proxmox / k8s cluster behind Authentik SSO, and deliberately kept to one container + one SQLite file so you can run it anywhere you'd run a reverse proxy.

Pre-1.0 (0.2.0-pre.1). Breaking changes still possible. The core (sessions, events, dashboard, hooks) is stable and used in production daily; the AI control-plane features (watcher, Ask, Telegram) are gated behind Labs flags that default off for exactly that reason.

Shipping cadence is high — see the CHANGELOG for what's landed recently.

• Try it — Getting Started walks you through the one-liner install.
• Contribute — 10 starter issues are tagged good first issue with file paths + acceptance criteria. New LLM adapters, new notification channels, CLI tools, a11y work.
• Ask — open an issue with the question label, or message me on GitHub.