Open research,built together.
2vane is the open-source lab of Tessovix. We build AI safety and AI tooling projects in the open, and invite anyone to read the code, run it, question it and improve it.
- Open source
- AI safety
- Apache-2.0
- Built with the community
- Part of Tessovix
01
Projects
- A
LoyalLens
- What
- White-box detection of secret loyalties in language models: a model covertly steered to serve one party, which behavioural audits miss because it acts on the loyalty but never says it.
- How
- Probes the residual stream against a content-matched control. Runs without a GPU on synthetic activations.
- B
V-Sentinel
- What
- A dual-control guardrail for Vietnamese public-service, education and healthcare chatbots
- How
- Deterministic rules plus an LLM classifier; fails closed; resists Vietnamese-specific evasion and redacts Vietnamese PII
- C
Graded Memory
- What
- Captures an organization's prompts, workflows and agent configs, and grades each KEEP, REVISE or RETIRE
- How
- Local secret and PII scan, redaction before any model call, append-only audit log
02
Contribute
Pick a project
Read the README, run the tests and try the demo. Every project runs locally without paid API keys.
Open an issue
Report a bug, question a result or propose an idea on the project's GitHub page.
Send a pull request
Small, focused changes with a test are the fastest to review. All projects are Apache-2.0.
03
About
2vane is part of Tessovix, the company behind Beaver. Tessovix product repositories live under github.com/tessovix; community projects live here, under github.com/2vane.
- Codegithub.com/2vane
- LicenseApache-2.0
- Contactsupport@tessovix.com