Quesen answers one question before every high-risk agent action: should this proceed? You describe the attempted operation as a typed security context — subject, action, target, tool (requested vs granted scopes), data classes + egress destination, and provenance — and a deterministic checker returns PASS / REVIEW / BLOCK / SKIP with reason codes and a replayable receipt (SHA-256 over the canonical input + the pinned ruleset commit). No model inference in the decision path, so the same input always yields the same verdict.
Canonical example: an agent tricked into POSTing a secret to an untrusted endpoint returns BLOCK (EGRESS_SECRET_UNTRUSTED); a benign public egress returns PASS. An unattested client-asserted payment grant returns REVIEW (UNVERIFIED_GRANT).
Works over direct HTTP (POST /validate, POST /tsc/validate), as a native MCP server, and via official SDKs for Python, TypeScript, LangChain, CrewAI and AutoGen.
pip install quesen-sdk # Python
npm i quesen-sdk # JavaScript / TypeScript
pip install quesen-langchain # LangChain
pip install quesen-crewai # CrewAI
pip install quesen-autogen # AutoGenFree sandbox key (no signup, no card): curl -X POST https://web-production-aa5ba.up.railway.app/sandbox/keys