Krisk documentation
July 19, 2026 ยท View on GitHub
Krisk has two independent entry points that can be combined when useful:
- chart library: pass Python data directly to Krisk in a notebook or script;
- research server: let an LLM investigate named read-only sources through MCP.
The chart library does not start a server or write to Krisk's metadata database. The research server uses SQLite locally unless the operator explicitly configures shared PostgreSQL metadata.
Start here
| Goal | Guide |
|---|---|
| Make a chart from pandas or Ibis | Typed charts |
| Try the complete local workflow | End-to-end tutorial |
| Configure an LLM or API client | MCP and REST |
| Preserve a conclusion and monitor it | Living research |
| Understand data access and persistence | Security and storage |
Notebook tours
The repository notebooks are executable, server-free walkthroughs with repository-local data and no undeclared visualization dependencies:
- Quick start
- Declarative visualization
- Legends, titles, and toolbox controls
- Reproducible charts and live refresh
- Themes and colors
- Tidy data plotting
- Waterfall and combination charts
The data lifecycle
flowchart LR
A["Named read-only source"] --> B["Saved query + fingerprint"]
B --> C["Typed chart"]
C --> D["Immutable Parquet snapshot"]
B --> E["Research methodology"]
D --> E
E --> F["HTML / Jupyter export"]
A -. "Live refresh" .-> F
An export contains the historical snapshot. Live data is requested separately and is never written over that evidence.
Version status
The current version is 0.9.0, an unpublished local-first beta. 1.0.0 has never
been published. The 1.0 implementation review remains in this repository as a record
of the pre-release audit and the reasoning behind the beta version.