AgentAssertRuntime Behavioral Contracts for AI Agents
Define rules in YAML and check structured agent state in Python with AgentAssert. The application supplies the signals and places checks before a protected action. Start with a synthetic approval-flag proof, then explore the research assumptions and enforcement adapters.
A prompt needs an observable check
An agent's claim of compliance is not runtime evidence. Define an invariant, pass the signals your application can measure, and check them before proceeding. Coverage depends on the adapter and action boundary you choose.
Key Capabilities
Hard and Soft Constraints
Separate inviolable safety boundaries from aspirational quality targets. Hard constraints halt execution on violation; soft constraints degrade gracefully and log for review.
Real-Time Drift Detection
Continuously monitors agent behavior against its contract during execution. Detects semantic drift before it compounds into a catastrophic failure downstream.
Reliability Scoring
Produces a single composite reliability score (Θ) per agent per session. Enables objective comparison across model providers, prompt versions, and deployment configurations.
Multi-Agent Pipeline Contracts
Compose individual agent contracts into pipeline-level guarantees. When Agent A hands off to Agent B, the contract enforces interface-level expectations at the boundary.
Enterprise-Ready Compliance
Designed with the EU AI Act in mind. Provides the audit trail, constraint documentation, and runtime evidence that regulators and compliance teams require for high-risk AI systems.
The synthetic proof checks a caller-supplied boolean: true passes; false raises ContractBreachError. It is not a PII classifier, accuracy benchmark or blanket guarantee. The arXiv preprints state the assumptions and protocols for research results.
Get Started
SkillFortify
Supply Chain Security for AI Agent Skills
Static analysis, behavioral sandboxing, and cryptographic attestation for the AI agent skill ecosystem. Detects malicious skills before they execute.
AgentAssay
Regression Testing for Non-Deterministic AI Agents
Token-efficient stochastic behavioral testing framework purpose-built for non-deterministic AI agent workflows. Part of the Qualixar AI Reliability Engineering platform.
SuperLocalMemory
Local-first Memory for AI Agents
Store and recall project context across AI agent sessions through MCP or CLI, with workspace isolation, explicit operating modes and auditable retrieval.