Skip to content
Krv Labs
Krv Labs · Open Source Registry
Full Parity with docs.krv.aigithub.com/Krv-Labs ↗

00 / Open-Source Stack

Scientific software built in the open.

Our open-source tools span three pillars: structural code quality, representation learning for complex tabular and clinical cohorts, and using AI safely in healthcare.

Open docs.krv.ai
01 · Code Quality (Topos) ↓02 · Representation Learning (Pulsar · Phil · TRAILED) ↓03 · Use AI Safely in Healthcare (Pasteur · Phirewall) ↓

Topos

Code quality.

Mathematical dependency-graph invariants and structural complexity metrics for agent-written and human-written software.

Plate 01 · Topos Dependency LatticeRUST + PYTHON · CLI & MCP

Topos

docs.krv.ai/topos ↗GitHub ↗

Structural code quality metrics for agent-written programs.

Topos treats software dependency graphs as mathematical structures, computing spectral, topological, and coupling invariants in milliseconds so autonomous coding agents and engineers catch architectural drift before it merges.

  • Dependency-graph invariants, cycle detection & spectral complexity scoring
  • Native CLI + MCP server for real-time refactoring guardrails in autonomous coding loops
  • Zero-config evaluation across Python, TypeScript, and Rust repositories
Explore Topos Overview

Pulsar · Phil · TRAILED

Representation learning.

Topological and multiverse methods that learn geometric structure across scales—clustering heterogeneous populations, regularizing EHR representations, and imputing missing data without collapsing rare phenotypes.

Plate 02PYTHON + RUST · MULTIVERSE TDA

Pulsar

Docs ↗GitHub ↗

Blazing-fast multiverse data exploration — based on our Nature Energy 2025 publication.

Pulsar implements our Thema multiverse clustering algorithm in Rust with Python bindings and an interactive web explorer. Instead of forcing one arbitrary clustering cutoff, Pulsar sweeps the entire resolution hierarchy to surface stable sub-populations.

  • Rust-accelerated hierarchical & topological multiverse clustering
  • Interactive browser UI + Python API + MCP server integration
  • Published methodology in Nature Energy (2025)
Explore Pulsar Overview
Plate 03PYTHON · TOPOLOGICAL IMPUTATION

Phil

Docs ↗GitHub ↗

Fill your data. A Multiverse Imputation Method powered by Topology.

Standard clinical imputation collapses patient variation into a single mean or median guess. Phil uses topological representations from TRAILED across multiple neighborhood resolutions to preserve manifold structure when imputing missing tabular EHR values.

  • Drop-in scikit-learn compatible PhilTransformer pipeline step
  • Multiverse resolution sweep powered by Euler Characteristic Transforms
  • Built-in phil-mcp server for agent-assisted data preparation
Read Phil Docs
Plate 04PYTHON + RUST · DIFFERENTIABLE ECT

TRAILED

Docs ↗GitHub ↗

Topological Regularization and Integrity Learning for EHR Data.

TRAILED computes fast, differentiable Euler Characteristic Transforms (ECT) over structured tabular and longitudinal electronic health record datasets—acting as the geometric backbone for Phil and downstream phenotype models.

  • Differentiable Euler Characteristic Transform (ECT) layers & regularizers
  • First-class NumPy, Pandas, Polars, and scikit-learn interoperability
  • Rust core for high-throughput cohort-scale topological signatures
Read TRAILED Docs

Pasteur · Phirewall

Use AI safely in healthcare.

Clinical AI requires stress-tested models and zero exposure of raw patient identifiers. Pasteur and Phirewall enforce those boundaries before and during analysis.

BSD-3-CLAUSE · CLINICAL AI STRESS-TESTING05

Pasteur

pypasteur / pasteur-cli
Overview →Docs ↗

“Open-source clinical AI stress-testing toolkit.”

Named after Louis Pasteur, Pasteur stress-tests clinical predictors under realistic deployment perturbations before they touch a hospital floor—simulating missing labs, noisy vitals, and label corruption.

  • Blackout (structured missingness), Jitter (sensor noise) & Flipper (label noise)
  • Dual Rust CLI (pasteur-cli) and Python library (pypasteur)
  • Automated robustness curves & clinical safety thresholds
SLATED RELEASE · LOCAL PHI VAULT + MCP GATEWAY06

Phirewall

phirewall + phirewall-mcp

“Use AI to analyze patient data, without the AI knowing who the patients are.”

Phirewall is a local vault and fail-closed MCP gateway. Clinicians register datasets and mark PHI columns in a human-only desktop interface; AI agents only ever interact with deterministic Tokenized Twins and small-cell-checked aggregate outputs.

  • OS-keychain encrypted local SQLite vault (phirewall.db) that never leaves the machine
  • Deterministic Tokenized Twins (e.g., PAT_000001) preserving longitudinal joins
  • Fail-closed k >= 5 small-cell suppression and automatic PHI token interception
InstallComing soon