King Hippopotamus

King Hippopotamus

King Hippopotamus is an independent software house: one operator, and the agent staff he architects and runs — with written charters, scoped write access, and gates that can refuse.

It builds the data platforms, agent operations and security layer behind its own financial research, and publishes the tools that came out of that work as packages anyone can install.

  • Builds data platforms
  • Runs multi-agent operations
  • Ships installable packages
  • Operates the security layer

Open to full-time engineering work where this kind of system is needed — remote or hybrid. Short, scoped engagements too.

Installable

Five packages on public registries. A nightly check reinstalls every one of them from its registry: the Python packages get their command line run and their module imported, the browser elements get their entry point resolved. The three Python packages were re-released after that check caught a breaking upstream change.

pip install IronDome

3.2.2PyPIGPL-3.0

CLI irondome

IronDome →

pip install golem-3dmcp

0.2.2PyPIMIT

CLI golem

golem-3dmcp →

pip install arelle-mcp

2.1.1PyPIApache-2.0

CLI arelle-mcp

arelle-mcp →

npm install md-spawn

1.0.0npmMIT

<md-spawn> element

md-spawn →

npm install sst-spawn

2.0.0npmMIT

<sst-spawn> element

sst-spawn →

What the house does

Three things the house does, each with one number behind it and a command that produced the number.

Acquire and govern financial data at scale

Collectors pull filings, prices, holdings and earnings calls out of unstructured upstream sources into a governed multi-database estate, where the quality gate refuses a bad row at write time instead of writing it and reporting later.

397,679,393rows under governance, across 28 named surfaces

Run agent work that verifies itself

Every stage gets a written charter, a scoped write path and a tool-call budget, and an independent stage re-runs the commands from a fresh context before any result is called done.

5executable gates in a research run, each returning an exit status

Operate one host, and watch it

Timers and cron carry the schedule, chain guards and watchdogs turn drift into a number, and a standard-library sensor scores 10 layers of the host every five minutes.

54scheduled jobs — 43 timer units and 11 cron lines

How I work

Charter

Every programme starts as a written architecture: objectives, a census of sources, the protocol, and the acceptance gates. The organisation of agents, schedules and checks is designed before it is built.

Clock

Pipelines run on a schedule and report to me, not the other way round. Chain guards, watchdogs and a host sensor turn drift into a number; decisions are logged before a change, not after.

Gate

Nothing ships on a claim. Numbers are re-derived from source, releases are gated by tests, and an independent check re-runs the evidence before anything is published — including this site.

Contact