This is direction, not commitments or dates. If you want to influence what gets built — or learn what others are working on — join the conversation in GitHub Discussions.
Intend to do
- Broader connector coverage. Expand the range of supported services, regions, and resource types across the clouds and platforms already connected (AWS, GCP, Azure, Kubernetes, GitHub, GitLab).
- Improved verification quality. Reduce false findings and increase the accuracy and completeness of the
verify_findingspass. - Expanded evaluation datasets. Grow the suite of evaluation datasets used to measure agent quality and catch regressions.
- Agent memory across sessions. Allow agents to carry context and findings forward from one session to the next.
- MCP support. Add support for the Model Context Protocol so Cynative can interoperate with MCP-compatible tooling.
- GitHub Actions integration. Ship a GitHub Action that runs agents as pull-request checks, making it straightforward to wire security research into CI.
Do not intend to do
- No hosted SaaS. Cynative runs in your environment. Your data, your credentials, and your findings never leave your infrastructure.
- No write-by-default mode. Read-only is and will remain the default. The action gate and network layer enforce this; no future version will flip that default.
- No telemetry or phone-home. The binary makes no outbound calls of its own. There is no usage reporting, no crash reporting, and no anonymous analytics.
- No model lock-in. Bring your own model (BYOM) is a first-class property. Even once Cynative ships a specialized model, every supported LLM provider will remain available.