Docs
Documentation
These docs cover the public InvarLock GitHub repository and OSS engine. Start with installation and the first evaluation loop, then move into artifact inspection, reference material, and assurance notes as your workflow matures.
If you want to discuss the private on-prem offering that is being refined with design partners, use the design-partners page instead of treating these OSS docs as product packaging.
Browse docs
Start here
Follow the same onboarding path as the upstream README: install the package, run a first evaluation, then move into baseline-vs-subject workflows and evidence review.
01
Getting Started
Environment setup, installation, and the first evaluation loop.
02
Quickstart
CLI highlights for the common workflows and the first artifact outputs.
03
Compare & evaluate (BYOE)
Run baseline vs subject comparisons with pinned pairing and guard checks.
04
Primary Metric Smoke
Tiny examples for ppl and accuracy paths before a larger evaluation run.
Quickstart commands
Start with the core package, add the Hugging Face extra only when needed, then run a first baseline-vs-subject evaluation.
pip install invarlock
# Optional: only for evaluate with Hugging Face-backed models
pip install "invarlock[hf]"INVARLOCK_ALLOW_NETWORK=1 INVARLOCK_DEDUP_TEXTS=1 invarlock evaluate \
--baseline <BASELINE_MODEL> \
--subject <SUBJECT_MODEL> \
--adapter auto \
--profile ci \
--report-out reports/evalChoose a path
Run the quickstart if you want to execute the CLI immediately, or inspect artifacts first if you want to understand the evidence model before running anything. Use the design-partner page only for questions about the private on-prem path.
Start with the common install and first-run path.
Review real reports before running the CLI in your own environment.
Use this only if your question is about the private on-prem path.
Synced from public upstream
Source ref
v0.8.0 (release tag)
Source commit
9119404
Synced
Apr 24, 2026, 6:48 PM
Docs / diagrams
51 docs, 30 diagram replacements
Artifact Trail
Start here if you need to understand what the evaluation produces before running the CLI in your own environment.
Example Reports
Inspect representative evaluation outputs and reviewer-facing attachments.
Reading a report
Understand PASS/FAIL status, paired metrics, provenance, and verification fields.
Browse
Start with quickstartUser Guide
13Core workflows, evaluation reports, evidence packs, and practical guidance for running evaluations.
Reference
17CLI flags, configuration, and API references you can bookmark.
Assurance
15Safety case and assurance artifacts: evidence, analysis, and verification posture.
Security
5Security model, hardening notes, and operational guidance.