Documentation

Cognitive integrity foragent runtimes.

Sentisec sits between your agent and the model provider. It scores every call, ships signed forensic evidence, and stops hijacked sessions before they reach your tools. Drop-in for the providers you already use.

Quickstart How it works

Install, then point an agent at it

Point your agent at Sentisec.

Keep your SDK and your agent code. Change the base URL, and every call is scored on the way through.

Full setup

curl -X POST https://api.sentisec.ch/ws/<your-workspace-id>/v1/messages \
  -H "Authorization: Bearer $SENTISEC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{"model": "claude-sonnet-4", "max_tokens": 256,
       "messages": [{"role": "user", "content": "hello"}]}'

Choose your path.

Quickstart

Point your agent at Sentisec and get your first verdict. One base URL, no code changes.

Start

Providers

Anthropic, OpenAI, Kimi, Google Gemini and Vertex AI. Connect a credential and calls dispatch immediately.

Browse

Internals access

Evaluating the full mechanism, or running the stack yourself? The protected documentation covers it.

Request access

See it work.

The problem

Why an agent follows an instruction it found in a web page — and why every action it then takes is one you already authorised.

The record

The sessions your agents ran, the verdict each action got, and the signed record behind anything elevated.

Wire it in, watch the verdicts land.

The public docs cover the drop-in path and the operator surface. The full mechanism and self-hosting are documented in the protected tier.

Request internals access Read the docs