Continuous security testing that runs inside your own network.
Trinetra tests the systems you own on a schedule and at every deployment. Each finding points to the exact file and line, or host and port, with why it matters, and Trinetra can draft the fix as a pull request. It runs on your infrastructure: the only connection out is to a language model, and with a model you host there is none.
Coming soon. On-prem, for teams that can't send code or traffic to a hosted scanner.
Every change, checked where it runs
A periodic audit is out of date the day after it lands. Trinetra keeps assessing, on your own machines, every time something changes.
Connect
Read-only access to your repositories and running services. Trinetra maps what you have into an asset graph and keeps it current.
Trigger
Scheduled sweeps, plus a step in CI/CD that re-checks every deployment. Each trigger is checked against your signed scope first.
Assess and pinpoint
Checks run only against assets in scope. A finding names the file and line, or host and port, explains why, and shows what else it touches.
trinetra trigger deploy
Fix and verify
For a vulnerable dependency it opens a pull request on its own branch, then re-runs the same check to prove the fix worked.
What Trinetra will not do
- Act as an attack tool. It assesses only systems you own and authorise, from inside your own network.
- Test outside your scope. Every target must be in a scope file signed with your key. Anything else is refused and logged, never touched.
- Start on an unsigned scope. The signature is checked at startup. An unsigned or forged scope stops it before any check runs.
- Reach where it shouldn't. Private, loopback and cloud-metadata addresses are blocked by default, and each connection is pinned to the address that was approved.
- Promote a fix on its own say. Fixes arrive as pull requests. Your autonomy policy decides how far it may go, and nothing moves until the check re-runs clean.
- Keep a record you can't check. Every action goes into a hash-chained audit log, and one command verifies the chain is intact.
What it checks today
Six detection engines, run by an agent that plans, acts, looks at what it found and plans again. Findings are deduplicated and tracked across runs, so a regression shows up as one.
- Dependencies
- Known vulnerabilities in the packages you ship, from the OSV database.
- Committed secrets
- Keys and tokens in your code, found offline and shown redacted.
- Exposed files
.envfiles,.gitfolders and cloud credentials served where they shouldn't be.- Security headers
- Missing or weak headers on every service in scope.
- Cookie flags
- Session cookies without Secure, HttpOnly or SameSite.
- Surface discovery
- What is actually running, added to the asset graph with its coverage.
For teams that can't use a hosted scanner
Hosted scanners need your code, your traffic or your list of assets. Trinetra needs none of them to leave your network.
- Regulated
- Fintech, health and government teams that can't send source, traffic or asset lists to a third party.
- Air-gapped
- Works with any OpenAI-compatible model endpoint, including one you host, so nothing has to leave the network.
- Security teams
- A standing assessment on your own infrastructure instead of a consultant once a quarter.
- Deployment
- One program and a Postgres database, on a server or in Docker on your internal network.