v1.4.0¶
Release date: 2026-03-24
v1.4.0 turns the SDK into a self-contained distribution by bundling MDMP directly into the package and hardening the surrounding safety and physiology paths.
Highlights¶
- MDMP now ships inside the SDK:
mdmp_coremdmp_aimdmp_flavorsmdmp_integrations- AI and demo workflows no longer depend on a separate MDMP repository checkout
- the combined distribution now carries Apache 2.0 licensing metadata and a
NOTICEfile for the bundled protocol code - safety and biology hardening from the latest audit is included in the same public release
What changed¶
Bundled MDMP runtime¶
The SDK now includes the MDMP implementation directly in src/, together with:
- conformance vectors
- bundled public key material
- MDMP CLI entrypoint support
- internal backend wiring for AI guard and artifact preparation
That means a normal SDK install can now provide MDMP-backed flows without needing a second repository on disk.
Safer default behavior¶
This release also ships the recent hardening work:
- corrected CGM metric boundary handling at exact values like
54,70,180, and250 - tightened URL validation for local Ollama and registry download paths
- safer model loading in research and learning components
- more consistent T1D-oriented physiological defaults and broad-bound validation
Documentation¶
The guides now better explain:
- how to install and update the SDK cleanly
- how to set up Ollama and connect it to the SDK
- where the medical and technical evidence comes from
- how bundled MDMP changes the old dual-repo workflow
Install¶
Latest release:
python -m pip install -U "iints-sdk-python35[mdmp]==1.4.0"
Then verify:
iints doctor
iints ai local-check --model ministral-3:3b
Why this release matters¶
Before v1.4.0, the SDK could integrate with MDMP but still relied on a separate protocol repository or external package path in some real-world workflows.
With v1.4.0, the SDK is much easier to move between machines, demonstrate live, and maintain as a single public package.