# Krim > KrimOS is the agent operating system for safe lending operations: AI co-workers whose every action is validated before it executes (Krim-Nyāya, pre-execution validation) and that learn from every recorded outcome (Krim-Learn), all inside the institution's own perimeter. Krim is building toward the full lending stack, including a safe, validated AI underwriter, through Kovida — the world lending model. Tagline: "The AI your regulator can read." ## Platform (KrimOS) - [KrimOS](https://www.krim.ai/krimos): the agent operating system for safe lending operations — one stack, from the world lending model at the foundation to the apps your teams and customers use. - [Kendra](https://www.krim.ai/krimos/kendra): the engine and the validation gate — Krim-Nyāya clears every action before it fires; Krim-Learn learns from every outcome. - [Kriya](https://www.krim.ai/krimos/kriya): the vocabulary — 500+ validated, credit-native action primitives across 20+ domains. - [Karta](https://www.krim.ai/krimos/karta): the AI co-workers, composed from validated primitives. - [Kupa & Kula](https://www.krim.ai/krimos/kupa): the command center for your teams — direct the work in plain language; supervise, configure and audit it from one pane. - [Krimkar & Kira](https://www.krim.ai/krimkar): the app your customers hold, and the AI advisor inside it — one relationship across every channel. ## Research - [Epistemic AI](https://www.krim.ai/epistemic-ai): the category Krim defines — AI whose every action is validated before it fires and whose reasoning an auditor can read end to end. Built on Navya-Nyāya formal logic. - [Kovida — the world lending model](https://www.krim.ai/research/world-lending-model): a complete, safe world model of lending with an agent harness — borrowers, products, markets, rules and the whole lifecycle — that agents reason and plan against across the full stack, from origination to a validated AI underwriter to collections. - [Safe Agent Harness](https://www.krim.ai/research/safe-agent-harness): the operational control layer that wraps AI agents — constraining what they can do, gating what they propose, and keeping a human in command. - [Research](https://www.krim.ai/research): the work under the product — making judgment machine-checkable, learning the operation, and pre-execution validation. ## Who it's for - [Lending](https://www.krim.ai/lending): the first and primary domain. - [Government](https://www.krim.ai/government): capability for public-sector operations. - [Enterprise](https://www.krim.ai/enterprise): one estate, one standard — for large institutions and smaller teams alike. ## Writing - [The EU AI Act's High-Risk Clock](https://www.krim.ai/insights/the-eu-ai-act-high-risk-clock): credit scoring is high-risk under the EU AI Act; Articles 10, 12 and 14 read as engineering requirements. - [Your Compliance Team Is Right to Say No](https://www.krim.ai/insights/your-compliance-team-is-right): the question that kills AI pilots is the right question — how do we know the action was allowed? - [Collections Is a Sequence Problem](https://www.krim.ai/insights/collections-is-a-sequence-problem): cure comes from a sequence of contacts; the unit of liability is the individual touch. - [Explain the Decision, Not the Model](https://www.krim.ai/insights/explain-the-decision-not-the-model): regulators never asked you to open the black box; they asked you to explain the decision. - [What Epistemic AI Means](https://www.krim.ai/insights/what-epistemic-ai-means): generative AI is trained to be plausible, agentic AI to act; epistemic AI is the missing third — validated before it fires, its reasoning legible end to end. - [The World-Model Moment](https://www.krim.ai/insights/the-world-model-moment): AI's frontier is moving from predicting the next word to predicting the next state of a world. A lending operation is a world too — where is its record? - [RBI Has Set the Bar for AI Lending. Almost No One Can Clear It.](https://www.krim.ai/insights/rbi-model-risk-management-2026-ai-lending): RBI's 2026 Model Risk Management draft quietly rewrites the rules for AI in lending. - [The automation gap](https://www.krim.ai/insights/the-automation-gap): why the most consequential work still runs by hand. - [The cost of being wrong](https://www.krim.ai/insights/the-cost-of-being-wrong): what a non-compliant action costs a lender. - [The credit gap is an operations problem](https://www.krim.ai/insights/the-credit-gap-is-an-operations-problem). - [The Model Nobody Can Validate](https://www.krim.ai/insights/the-model-nobody-can-validate): RBI's model-risk draft asks for validation almost no AI vendor can evidence — and what real validation looks like. - [Audit after the fact is a confession](https://www.krim.ai/insights/audit-after-the-fact-is-a-confession): the case for pre-execution validation. - [Sovereignty is not optional](https://www.krim.ai/insights/sovereignty-is-not-optional): why regulated AI must run inside the perimeter. ## More - [Architecture](https://www.krim.ai/architecture): how KrimOS fits the stack you already run. - [Trust](https://www.krim.ai/trust): the safety, sovereignty and audit posture. - [Services](https://www.krim.ai/services): deep-dive, proof on your own data, then production — proof before you commit, at every stage. - [Company](https://www.krim.ai/company) · [Contact](https://www.krim.ai/contact) ## Notes - Krim does not publish customer names or deployment metrics. Government framing is capability, not a claimed track record. - Kovida and the AI underwriter are an active area of Krim research and engineering.