CoreKit#
Your project starts from working software, not from plumbing.
CoreKit is the open-source Python toolkit I bring to every engagement. It lands in your repository on day one with data lakes, models, agents, pipelines, web tools, deployment and documents already wired, so the budget goes to the pipeline, model or agent that has to ship. When the engagement ends, the code stays with your team under Apache 2.0.
Book a discovery call Consulting experience
How it works in an engagement#
-
I lead the work
I own delivery of a defined pipeline, model or agentic workflow on CoreKit: scope, implementation, operations dashboard and the software your team inherits.
-
Your team ships faster
I work inside your applied science or AI engineering team, with scaffolding, pipelines and agent plumbing already in place, so the team spends its time on the problem.
-
It extends what you run
I add data lakes, pipelines, models or agents to a codebase you already operate, on the platform you already run.
-
You keep it
CoreKit is one Python wheel in
.vendor/corekit/, wired as a local dependency, with the complete source inside. Your engineers and coding agents can read it, extend it and keep running it after I leave.
What's inside#
31 packages in seven areas, each a typed and tested Python package with its own command line or web interface where it needs one. The launchpad puts the web tools behind one login, and the sandbox wraps everything that runs commands on behalf of an agent or a user. Select a package to read what it does.
Data and machine learning
Pipelines and deployment
every package builds on
Foundation
Data and machine learning#
From raw files to model results you can compare across runs.
Typed database access
corekit.db
Queries application data with SQLModel and analytics with DuckDB into Polars, and lets only read-only queries through where required.
Analytical data lakes
corekit.db.lakehouse
Loads CSV, JSONL and Excel files into a DuckDB lake and records where each table came from and how the tables relate.
Time series collections
corekit.dataseries
Holds many series in one table, with chainable transforms, loaders and fingerprint plots to compare them at a glance.
Machine learning presets
corekit.ml
Runs scikit-learn and XGBoost grids and reports ranking metrics with bootstrap confidence intervals, so every run is measured the same way.
Figures and animations
corekit.plot
Draws figures with project defaults from one context manager and renders animations to GIF or MP4.
Safe artifact loading
corekit.safepickle
Checks pickles against an allow-list before loading, so cached and handed-over artifacts load as data, never as code.
AI agents#
Agents with scoped tools, hard limits and a sandbox around them, and results you can trace to their sources.
Agent runtime
corekit.agent
Runs Pydantic AI agents with scoped file, SQL, code and web tools and usage limits, as a step of the pipeline.
Claude Code launcher
corekit.claude
Starts Claude Code with the project's settings and packaged prompts, inside the sandbox.
OS-level sandbox
corekit.sandbox
Enforces one policy for paths, network and terminals, with Seatbelt on macOS and bubblewrap on Linux.
Evidence-backed research tables
corekit.tabulate
Has agents build tables from public sources, where every cell carries the evidence behind it.
Pipelines and deployment#
Work that runs the same way every time, that stakeholders can follow, and that deploys from the repository.
YAML pipelines
corekit.pipeline
Runs pipelines defined in YAML, with a task registry, layered overrides and a reproducible directory per run.
Operations dashboard
corekit.pipeline.web
Lets stakeholders start runs, follow logs, preview artifacts and query results from a browser.
Deployment from the repository
corekit.shipyard
Deploys and provisions hosts over SSH from one manifest per service, with a dry run first. This website ships with it.
Host monitoring
corekit.sysmon
Shows CPU, memory, disk, network and load on a live page, so a slow run is explained by numbers from the machine.
Web tools and access#
Internal tools with sign-in and permissions in place from the first day.
Web services
corekit.web
Serves FastAPI apps with production-safe defaults, started the same way in every project.
Users and permissions
corekit.identity
Manages users, organizations, permissions, tokens and API keys, from the browser or the command line.
One login for every tool
corekit.launchpad
Signs people in once and shows each of them the web tools they are granted.
Browser terminal
corekit.terminal
Runs a command-line tool in a browser tab, sandboxed, for colleagues who never open a terminal.
Documents#
Reports and decks generated and checked in the same repository as the code they describe.
Documents and decks from source
corekit.authoring
Builds PDF and DOCX from Markdown and PPTX decks from Python, with packaged templates.
Document review
corekit.proofread
Flags broken links, inconsistent terms, leftover placeholders, tone and confidentiality issues before a document ships.
API documentation
corekit.doc
Generates a browsable API site from the code, with live reload.
Sealed files
corekit.seal
Seals a report into one HTML file that decrypts in the browser with a key, safe to send by email or chat.
Engineering and delivery#
The tooling that keeps the codebase reviewable, fast where it matters, and ready to hand over.
QA tooling
corekit.dev
Adds pytest fixtures, version bumping and a checker for broken links, anchors, symlinks and YAML paths.
Git workflow
corekit.git
Keeps commits conventional and pull requests consistent across the team.
Rust for hot paths
corekit.oxidize
Rewrites measured Python hot paths in Rust with PyO3, with parity tests and benchmarked speedups.
Project scaffolding
corekit.scaffold
Starts a new uv workspace with the vendored CoreKit wheel and the same tooling already wired.
Distribution builds
corekit.utils
Merges selected packages into one versioned wheel, optionally encrypted, ready to vendor into your repository.
Disk cleanup
corekit.sweep
Finds stale virtual environments and caches, reports their size and moves them to a trash you can restore from.
Foundation#
What every other package builds on.
Typed core
corekit.core
Validates typed configuration with readable errors, sets up logging and caches results.
Command-line setup
corekit.cli
Gives every command-line tool the same options, logging and error handling.
Observability
corekit.observability
Traces each run with OpenTelemetry spans, durations and host stats.
Day one in your repository#
The wheel is already a local path dependency, so one sync gives your team the toolkit, its documentation and its source:
# Install the project and CoreKit
uv sync
# Browse the API documentation with live reload
uv run corekit.doc serve
# Extract a working copy of the source
uv run corekit.doc extract-source --output corekit-source
Built on#
Python 3.14, uv, ruff, ty, pytest, Pydantic, Typer, FastAPI, DuckDB, Polars, SQLModel, scikit-learn, XGBoost, Pydantic AI, OpenTelemetry, Logfire, Zensical, and Rust with PyO3 and maturin. See the full technology stack on the Experience page.
Have a pipeline that has to ship?
CoreKit comes with every engagement. The quickest way to find out whether it fits your stack is a short call. Prefer email? Write to [email protected].