CAGE-lite v0.1.0 product preview
Pre-releaseCAGE-lite v0.1.0 product preview
CAGE-lite is an open-source reference implementation of the CAGE framework for prebind assurance of AI-agent actions.
Agent platforms govern how agents run. CAGE governs whether agent actions are allowed to become business consequences.
What this release includes
- CAGE Warrant schema
0.4 - durable boundary-decision, evidence, and effect records
HELD,ADMITTED, andNARROWEDboundary outcomes- durable
NO_BIND,BOUND, and narrowed-effect proof - replay linkage between an original held Warrant and an admitted replay
- independent Warrant digest verification
- modular Streamlit interface with:
- Overview
- Boundary Runs
- Warrants
- Replay
- Architecture
- public examples for held, approved, and narrowed payments
- GitHub Actions validation on Python 3.10 and Python 3.13
- packaged Streamlit application and CAGE logo
- contribution guidelines, security policy, and changelog
Demonstration scenario
The flagship demonstration proposes a USD 75,000 vendor payment by an agent whose standing limit is USD 50,000.
Without the required approval:
- boundary outcome:
HELD - effect disposition:
NO_BIND - protected API executed: no
- system of record written: no
After approval is supplied and the action is replayed:
- boundary outcome:
ADMITTED - effect disposition:
BOUND - protected API executed: yes
- system of record written: yes
- replay Warrant links to the original held Warrant
Version labels
- Product: CAGE-lite v1 product preview
- Python package:
0.1.0 - CAGE Warrant schema:
0.4
These version labels are intentionally separate.
Validation
- 95 tests passed
- CI passed on Python 3.10 and Python 3.13
- wheel and source-distribution builds passed
- clean wheel installation passed outside the source repository
- installed replay demonstration completed successfully
- packaged application assets, LICENSE, and NOTICE were verified
Scope
CAGE-lite does not replace agent runtimes, IAM, policy engines, guardrails, MCP or API gateways, observability platforms, agent evaluation frameworks, or hyperscaler governance services.
CAGE consumes signals from those systems and evaluates whether a proposed action may cross the business consequence boundary and become binding.
See CHANGELOG.md for the complete release summary.