HiveDoctor

The self-healing watchdog for Honeycomb.
Keeps your Honeycomb daemon alive, repairs it when it breaks, and reports home when it cannot.

      

_{A Legion Code × Activeloop collaboration · powered by Deep Lake}

---

Honeycomb runs a local daemon that gives your AI coding agents shared, persistent memory. HiveDoctor is the little bee that keeps that daemon healthy. It watches the daemon, fixes it the way a careful operator would, and when it genuinely cannot, it leaves a clear report and (unless you opt out) tells the maintainers, so problems get solved before they cost you a morning.

It is deliberately small. Zero runtime dependencies, Node built-ins only, every action wrapped so a failure is logged rather than fatal. The thing that watches your daemon is built to be even harder to kill than the daemon itself.

Installed automatically by the Honeycomb one-command installer. You usually never run it by hand. This page is for when you want to.

---

Why it exists

A daemon can fail in quiet, annoying ways: it wedges but still holds its port, it gets launched from a directory it cannot write to, a stale global install serves old routes, credentials go bad. When that happens on your machine, you see a broken product and the maintainers see nothing. HiveDoctor closes that gap on both ends: it repairs the common failures on the spot, and it gives the team real visibility into the ones it cannot, so support stops being guesswork.

What it does

• Watches the daemon. Probes http://127.0.0.1:3850/health on a fixed interval and reads the per-subsystem health detail, so it knows what is wrong, not just that something is.
• Heals automatically. Runs an escalating repair ladder with exponential backoff, the same moves a human operator would make, and goes quiet again the moment the daemon is healthy.
• Escalates loudly when it can't. If the ladder is exhausted, it writes a structured "needs attention" report, serves a tiny local status page, and (unless opted out) sends the diagnosis home.
• Keeps Honeycomb current. Checks for a blessed new release and updates the daemon safely (verify, then roll back on failure). On by default, easy to opt out.
• Survives anything. Supervised by your OS (launchd / systemd / Windows Scheduled Task), so it restarts on crash and comes back after a reboot, independently of the daemon it watches.

How it heals

When the daemon stops looking healthy, HiveDoctor climbs a ladder, backing off between rungs and stopping the instant health returns:

Rung	Action	When
1	Restart the daemon	first response to an unhealthy daemon
2	Reinstall the daemon	after 3 failed restarts (fixes a corrupted or stale global)
3	Remove a conflicting Hivemind	whenever a clashing @deeplake/hivemind global is detected (the package only, never your shared ~/.deeplake/ data)
4	Escalate	when nothing above worked: persist a report, surface it locally, and notify home

It will never delete your credentials. If it suspects a credential fault, it escalates instead of touching them.

Install

HiveDoctor ships with Honeycomb. The one-command installer sets it up and registers its OS service automatically. To skip it, pass --no-hivedoctor (or set HONEYCOMB_NO_HIVEDOCTOR=1) at install time.

To install or update it on its own:

npm install -g @legioncodeinc/hivedoctor
hivedoctor install-service   # register the OS service (restart-on-crash, start-on-boot)

Commands

Run hivedoctor with no arguments for the banner and menu. The ones you will actually use:

Command	What it does
hivedoctor status	daemon health, service state, versions, last heal, opt-out flags
hivedoctor diagnose	classify health and print the recommended fix, taking no action
hivedoctor heal	run the repair ladder once (disruptive rungs confirm first)
hivedoctor restart	restart the daemon (rung 1)
hivedoctor update [--check]	update the daemon via the safe blessed gate
hivedoctor logs	tail the local incident log
hivedoctor install-service / uninstall-service	register or remove the OS service
hivedoctor self-update	update HiveDoctor's own package (the only thing that does)

HiveDoctor never updates itself in the background. It is built not to need it; self-update is the single, explicit way to bump it.

Telemetry and privacy

By default HiveDoctor sends a small amount of operational telemetry to the maintainers so they can fix problems proactively: error events, a periodic installation-health snapshot, and the steps it attempted during a repair. It is sent as OpenTelemetry logs to PostHog.

It is built to be honest about this:

• What is sent: health state, daemon and HiveDoctor versions, OS, and the fact and outcome of a repair (an allow-list, deny-by-default).
• What is never sent: your credentials, tokens, file contents, or personal data.
• Opting out: set DO_NOT_TRACK=1 or HONEYCOMB_TELEMETRY=0, or toggle it in the Honeycomb dashboard. When opted out, nothing leaves your machine. Telemetry is always fire-and-forget and never blocks a repair.

Design principles

1. Incapable of crashing. Node built-ins only, zero runtime dependencies. Every probe and repair runs inside a try/catch that logs and continues, with a global last-resort net on top.
2. More reliable than what it watches. Supervised by the OS, never by the daemon. The two never depend on each other to stay alive.
3. Loopback and registry only. It reaches the daemon over 127.0.0.1 and the outside world only over npm and the telemetry sink. No new inbound ports.
4. Least blast radius. Destructive actions are scoped and reversible-by-design; a bad release cannot auto-propagate without passing the blessed-version gate.
5. Silent on the happy path, loud on the hard path. A healthy probe is a debug line; an unhealable install is a high-signal escalation.

Development

Self-contained: its own tsconfig.json and vitest.config.ts, independent of the repo-root gates.

cd hivedoctor
npm install        # dev deps only
npm run typecheck
npm run test
npm run build      # produces the single-file bin at bundle/cli.js

License

AGPL-3.0-or-later. A Legion Code × Activeloop collaboration, powered by Deep Lake.