Skip to content

v0.1.10 — metadata backend + a slimmer install

Released: 2026-07-21

Headline: tycoon gets dramatically smaller and more focused. The ingestion rewrite's first milestone lands — a typed MetadataBackend protocol behind all run history — and the FastAPI web UI, Dagster orchestration, and Nao/ask AI agent are removed outright, dropping the lockfile from 174 to 109 packages. The cycle plan is in docs/proposals/v0.1.10-scope.md; the rewrite's map is the new ingestion architecture doc.

Landed

  • MetadataBackend protocol + DuckDB file backend (#82, PRs #136/#137) — run history now flows through typed RunStarted / RunCompleted / RunFailed events and a structural Protocol instead of raw SQL against hardcoded table names. Every dlt dispatch emits lifecycle events (best-effort — a metadata write never blocks a run), and data history / data status read through the backend. Alternative backends (MotherDuck, Postgres, in-memory for tests) become an implementation of the Protocol, not a fork of the capture path.
  • Server / Dagster / ask extras removed (#139, PR #147) — ~15 modules deleted. tycoon ask and tycoon register llm are gone (typer's standard "no such command" error, no stubs). The init wizard is now 4 steps: ingestion → warehouse → dbt → rill. tycoon start/stop manage exactly two servers: Rill dashboards (:9009) and the Quack live warehouse (:9494).
  • Ingestion architecture doc (PR #149) — the plain-English map of where ingestion is going: Runtime / Destination / MetadataBackend protocols, the dlt tier model, and the milestone sequence for #81.

Upgrading from 0.1.9

  • If you used tycoon ask: the AI-agent surface is removed. Existing ask: blocks and stack.orchestrator* keys in tycoon.yml are ignored harmlessly — no migration needed, no file edits required. An AI-assisted surface will be rethought as part of the rewrite (#78).
  • If you used the Dagster integration: tycoon run <tool> still forwards to any CLI in your environment, so a self-managed Dagster (or Airflow, or cron via tycoon schedule) keeps working — tycoon just no longer manages it.
  • If you installed extras: pip install 'database-tycoon[server]', [dagster], and [ask] no longer exist. A plain pip install database-tycoon carries everything the CLI does.

Behind the scenes

  • Real test coverage rose from ~65% to ~73% with the deletion of the largely-untested server/orchestration code; the coverage floor moved to 68% with the same drift headroom as before.
  • pandas is now an explicit, pinned core dependency — nothing in tycoon imports it, but dlt's on-demand filesystem source needs it for CSV reading (optional-extra treatment tracked in #148).
  • All user-facing hints, docs pages, and wizard copy that referenced the removed surface were swept, and the CLI-surface stale-string sentinel now guards tycoon ask / register llm / Nao references so this class of drift fails tests.

What's next

The ingestion rewrite continues milestone by milestone — source catalog manifest, Runtime protocol implementations, and the dlt tier model. See the ingestion architecture doc and #81 for the sequence.