GitHubSign inStart testing

Product overview

One privacy loop, in three places.

Test what a run disclosed, stop it before it ships, and decide each call at runtime — with evidence someone else can check.

Create a workspace Run the quickstart
01

Testing

Find every leak in a run, on every channel.

  • Trace analysis
  • AgentRisk scoring
  • Encoded leaks
  • Adversarial red-team
Open 02

Build & CI

Catch it before it ships.

  • Static code scan
  • MCP preflight
  • CI policy gate
  • Agent API
Open 03

Runtime & evidence

Decide each call before it leaves, and prove it.

  • Runtime gateway
  • Model gateway (LLM proxy, LiteLLM)
  • Continuous monitoring
  • Public trust page
Open

Continue exploring

From explanation to implementation.

Security architectureSee where evidence runs, what is stored and how projects are isolated.Open resource Multi-agent privacyTrace disclosures across coordinators, specialists, memory and tools.Open resource Complete documentationInstall, instrument, score, red-team and enforce AgentLeak.Open resource