One business rule. Thousands of test conversations.
CIS reads your live Cognigy agent, builds a structural model of its logic, and drafts high-coverage test specs from that model — so coverage grows with your agent instead of your team's authoring time. Every generated spec passes semantic review and carries full provenance before it's marked ready.
How it works
From any agent, to assured quality.
Five stages, each producing evidence the next one is allowed to trust — from your agent's live logic to an approved test spec with a full paper trail behind it.
Source adapters
Reads your agent as it's actually built — live Cognigy flow definitions, not a re-typed description of your logic.
Canonical agent model
Normalizes what it reads into a structural model — journeys, decisions, modules, integrations, and target outcomes — a real model, not a black-box summary.
Intelligence engine
Graph, path, and dependency-impact analysis surface the decision points, branches, and flow interactions actually worth testing.
Test intelligence
Drafts test specs against coverage obligations, validates them structurally, and runs a semantic review pass with one bounded revision before anything is marked ready.
Provenance & evidence
Every generated spec carries model-invocation evidence and revision lineage, hashed and persisted — a generated test can be defended, not just trusted.
CIS is
A test-generation engine for Cognigy conversational agents, live today. It reads your agent's actual flow definitions, builds a structural model of the logic, and drafts test specs against real coverage obligations — every generated spec goes through structural validation and a semantic review pass, with one bounded revision, before it's marked ready to run.
It is not
- Multi-platform yet — Cognigy is the live source adapter today; LangGraph, LangChain, and RAG/knowledge-base ingestion are on the roadmap, not shipped.
- A replacement for the hand-written edge-case specs your team already trusts — CIS grows coverage alongside them, it doesn't retire manual authoring.
- A black box — every generated spec carries model-invocation evidence and revision lineage you can inspect, not just a pass/fail label.
Bring one business rule. See the coverage it generates.
We'll run CIS against one real rule from your live Cognigy agent and walk through every generated spec — the model, the coverage obligations, and the provenance behind it — with you.