An open-source-first, silo-tenant productionalization platform for quant shops.
Ship a model from research to audited production without an engineering hand-off. Answer your next LP operational due-diligence questionnaire from screenshots in the UI. Run in our managed GCP — or in your own, with BYOC.
Install pq, then start the platform locally and run a strategy end to end.
Supported on macOS, Linux, and Windows. pq takes care of everything else — if something
it needs is missing, it tells you.
pqmacOS or Linux:
curl -fsSL https://pyquant.io/install.sh | bashWindows (PowerShell):
irm https://pyquant.io/install.ps1 | iex
The installer puts the pq binary on your PATH and prints the version on
success. Re-run the same command any time to upgrade.
pq doctor # checks runtime prerequisites and port availability
Run this if pq up fails or before you start, to confirm your machine is ready. Skip
it on a known-good setup.
pq up # starts pyquant locally
pq new hello-world # scaffolds the strategy in the current directory
cd hello-world
pq run # trains the model and records the run pq up prints the URL of the platform UI when it finishes. Open it to see your strategy
registry, the run detail (metrics, parameters, model artefact, walk-forward evidence, PBO/DSR),
and the audit log.
Run pq down when you're finished.
Inside any strategy directory, pq e2e runs the full pre-push battery — lint, type check,
unit tests, and a containerised run — so you catch the same issues CI would before you push.
pq e2e # lint + type check + tests + containerised runHow we're different
Your data and code stay in your cloud.
Every tenant runs in a dedicated GCP project — optionally in your own GCP organisation. No shared database, no shared application process, no shared secrets vault. LP operational due-diligence questions about residency and isolation answer themselves.
No vendor lock-in — you can self-host if you choose.
Polars, Postgres + extensions, APScheduler, MLflow, FastAPI, React, Vite, uv, Cloud Run. Every load-bearing component is permissively licensed. A fund that wants to pull infrastructure in-house in three years can do so without a re-platforming exercise.
Walk-forward, PBO, DSR, bi-temporal data as defaults — not add-ons.
PBO/DSR computed on every backtest. Walk-forward validation gates model promotion. CPCV instead of vanilla k-fold. _knowable_at filtering at query-build time, not researcher discretion. These are quant-grade defaults — not optional add-ons you configure.
Evaluating for a specific fund? Email us.