# Rensei

> The deterministic runtime that makes agent-authored code defensible in production.

Rensei compiles declarative workflows to a durable execution graph. LLMs run as
bounded operators inside workflow steps; the control loop is deterministic code.
Cedar policy decisions run in the hot path. Every event lands on an Ed25519
hash-chained audit trail verifiable against per-workspace keys; the
verification protocol and key-discovery endpoints are published at
https://rensei.ai/.well-known/audit-keys.json. Decision provenance binds model
version, prompt, context, and policy ruling to every action.

Status, stated plainly: pre-revenue, design-partner phase. SOC 2 Type I is on the
roadmap and is not held. Three model providers run in production today: Anthropic,
OpenAI, and Google. Cloud deployment today; VPC and on-premises are on the
published roadmap. Policy enforcement is per-callsite and fails closed; there is
no separate network gateway appliance. Per-line provenance and survival
measurement are live; the survival-to-posterior wiring is in active integration.

The execution layer is open source: donmai (MIT), one Go binary.
Repo: https://github.com/RenseiAI/donmai
Install: brew install RenseiAI/homebrew-tap/donmai

## Pages
- https://rensei.ai/ : platform overview
- https://rensei.ai/platform : platform reference
- https://rensei.ai/open-source : the OSS boundary, stated exactly
- https://rensei.ai/security : security disclosure; certification status declared
- https://rensei.ai/evidence : the claims ledger; every claim carries a status
  (VERIFIED, MEASURED, IN INTEGRATION, ROADMAP, DECLARED) and a provenance class
- https://rensei.ai/pricing : engagement terms
- https://rensei.ai/manifesto : founder manifesto
- https://rensei.ai/docs : documentation