hosomaki

A lightweight
intelligence layer
for Linux

Reads your system and tells you what's happening in plain language. No cloud. No telemetry.

Get startedRead the docsGitHub
No cloudNo telemetryLocal · OllamaRead-onlyMPL-2.0
commands

Everything your system
needs to say.

One pipeline behind every command. Each run is sanitised before it reaches the model, and interpreted entirely on your machine.

hosomaki explainExplain anythingServices, boot logs, dmesg, piped streams, inline messages, running processes. If it logs, explain reads it.hosomaki statusSystem healthA snapshot of uptime, memory, disk, failed services, and recent errors.hosomaki doctorFull diagnosisDeeper than status. Surfaces concrete investigation steps for every identified issue.hosomaki watchReal-time tailTails a service journal and explains error bursts as they arrive.hosomaki auditChange detectionSnapshots your system state and diffs against it. Files, packages, listening ports, and unit states.hosomaki portsPort exposureLists every listening socket with process names and flags anything unexpected or worth investigating.hosomaki timersSystemd timersInspects all timers and flags any that have failed.hosomaki cronsCron jobsReads all crontab files and explains each job. Flags broken or suspicious entries.hosomaki mountsMount healthInspects active mounts, flags stale NFS, and warns on disks approaching capacity.hosomaki updatesPending updatesExplains what each pending package update changes before you apply it. Flags security fixes and reboot requirements.hosomaki historyPast resultsRevisit and summarise past results from explain, why, audit, status, and doctor without re-collecting system state.hosomaki firewallFirewall rulesExplain active firewall rules, with cross-checking against listening ports to flag rule mismatches or missing listeners safely.shell-integrationAuto-explain failuresPrefix any command with explain and failed commands get diagnosed automatically.
architecture

One pipeline.

Sanitisation is the mandatory first stage of every invocation. Nothing raw reaches the model.

  1. Collect

    Raw system data. Never stored, never modified.

  2. Sanitise

    IPs, hostnames, paths, UUIDs, and credentials stripped before the model sees anything.

  3. Prompt

    A tightly constrained prompt with a strict JSON schema.

  4. Validate & Repair

    Schema and semantic checks on every response.

  5. Render

    Typed, validated output formatted for the terminal. Raw model text never reaches the screen.

privacy

Stays on
your machine.

The model runs locally. No data leaves your system.

  • Local model via Ollama

    All inference runs on localhost. Works with any model Ollama supports.

  • Sanitisation before prompting

    IPs, hostnames, paths, UUIDs, and credentials stripped before the model sees anything.

  • Zero telemetry

    No analytics, no crash reports, no usage metrics. The binary makes no outbound connections except to your Ollama instance.

  • Strictly read-only

    Every command collects data and surfaces insights. None of them modify the system.

why Hosomaki exists

Your system has always known what's happening. Hosomaki just grew into the layer that reveals it.

Hosomaki began as a simple idea. It started as a tool that could translate logs into something mortals like you and me could read without digging through journal entries or kernel messages. But the deeper the project went, the clearer it became that translation wasn't the real goal. Linux already speaks, constantly. The challenge is making sense of everything it says.

That's where Hosomaki evolved. It no longer just interprets logs. It builds a structured understanding of your system by correlating signals, validating assumptions, and explaining what matters and why. It reads the same sources you would, strips anything sensitive, and produces a grounded analysis. Not a chatbot reply, but a deterministic pipeline with schema checks and repair loops before anything reaches your screen.

The model runs on your machine. The data stays on your machine. Your data belongs to you.

open source · mpl-2.0

Clone and run.

No account. No config file on first run. Requires Go 1.23+ and Ollama running locally.

license
Mozilla Public License 2.0
runtime
Linux · Go 1.23+
AI backend
Ollama (local, localhost)
telemetry
none, ever
  1. # install Ollama
  2. # pull a model (llama3.2:3b is the default)
  3. # clone the repo
  4. # build and install
Full install guide GitHub →