Kihan
ListedKihan provides constitutional computing and verifiable authorization to govern AI agent actions before they commit.
About
Kihan builds constitutional computing architectures separating AI proposals from the authority to commit actions. Its product, Inktomi, evaluates proposed agent calls against policy, routes exceptions for human authorization, and issues signed single-use permits verified by integrated tools. It provides tamper-evident logs and decision replay for compliance.
Detailed overview
Overview
Kihan is a software company that develops constitutional computing, an architecture designed to separate AI agent proposals from the authority to execute actions. Its flagship product, Inktomi, evaluates consequential agent requests against deterministic policies and issues verifiable, single-use permits that integrated tools must check before committing changes. The platform operates independently outside the language model, providing cryptographically verifiable proof and audit trails for automated decisions.
Key Features
- Single-use signed permits — Inktomi evaluates proposed calls and their specific arguments to issue signed, single-use permits that downstream tools verify before executing an action.
- Tool-level enforcement — The architecture binds authorization checks directly to the integrated tool rather than relying solely on network proxies, refusing direct calls that lack valid permits.
- Mathematically specified precedence — Policies operate under formal mathematical specifications where rule precedence is strictly defined, ensuring deterministic outcomes independent of evaluation order.
- Replay and re-derivation — The system records inputs alongside signed rule versions in tamper-evident logs, enabling auditors to re-derive and verify historical decisions.
- Human-in-the-loop exception routing — Proposals exceeding delegated thresholds are automatically routed to designated human authorizers authenticated via existing identity providers.
- Centralized governance console — Operators can inspect agent reach, track call history by owner, evaluate refused actions, and view geographical data destinations.
Who It's For
Kihan targets enterprise engineering, compliance, and security teams deploying autonomous agents in regulated sectors such as financial services, healthcare, legal, insurance, and government. It is built for organizations managing consequential workflows like financial transfers, claims processing, and sensitive record modifications. The platform also assists enterprises preparing for governance frameworks, including high-risk human oversight requirements under the EU AI Act.
Notable Strengths
Inktomi demonstrates rapid policy evaluation, recording a 1.22 ms median governance decision latency in Kihan's benchmark harness. The platform supports flexible enterprise integration through self-hosted deployments, standard SIEM event streaming, and compatibility with agent tooling including Claude Code, Codex, and Model Context Protocol (MCP). Furthermore, the underlying constitutional computing framework is grounded in formal mathematical research forthcoming in IEEE Computer.
Website link is available on the Verified plan
