Architecture
One runtime, under every action.
Run the whole lending stack on KrimOS, or layer it onto the systems you keep — origination to collections, every action validated.
The runtime
Run the whole stack, or layer onto yours.
Inside the runtime
The machinery that makes the guarantees hold.
The runtime layer is named Kendra. It is realised as eight engineering modules. Each does one job, and together they are why validated AI can act in regulated work.
Orchestration
Krim-Core
Routes each request to the right co-worker; runs sequential, parallel and hierarchical workflows; manages context, retries, fallbacks and human-in-the-loop; emits events.
Scheduler
Krim-Karya
Owns when work runs: timed and recurring tasks, deferred actions, retry and back-off timing, and time-window rules such as contact hours and regulatory deadlines, so every action fires only when it is permitted.
Knowledge base
Krim-Fabric
The per-jurisdiction regulatory calendar and rule set, the anonymised cross-tenant pattern library, and each tenant’s own policies and product rules.
Policy engine
Krim-Govern
A seven-level policy hierarchy encoding law, regulator guidance, internal rules and operational guardrails, enforced per tenant.
Validator
Krim-Nyāya
The staged pre-execution pipeline of 33 validators across three Navya-Nyāya families that gates every action before it fires.
Learning orchestrator
Krim-Learn
Coordinates ten learning loops across the workforce: outcome feedback, error and pattern analysis, temporal and semantic learning.
Metered record
Krim-Ledger
Logs every action immutably and meters it in Krim Work Units, serving the audit trail and usage-based billing from one source.
Telemetry
Krim-Sense
Metrics, logs, alerts and dashboards across the stack, feeding Kupa and internal monitoring.
The substrate
What the co-workers stand on.
Beneath the modules sits a shared substrate: the memory the workforce thinks with, the orchestration that keeps long work durable, and the single domain model they all speak.
Shared memory
Four tiers (working, short-term, long-term, episodic), shared across the workforce, so a pattern one co-worker learns is available to the rest without retraining.
Durable orchestration
Long-running workflows run on Temporal-class orchestration in Krim-Core, durable across crashes and deploys, with automatic retries, saga-pattern rollbacks, event sourcing and deterministic replay for incident investigation.
Shared domain model
One model of 15 core entities: Tenant, Account, Borrower, Loan, Interaction, Decision, PaymentPlan, Policy, ComplianceEvent, AuditLog and more. It is the common language every co-worker reads from and writes to.
The integration fabric
It fits the stack you already run.
40+ connectors span core banking, loan origination and servicing, CRM, telephony, messaging, payments, credit bureaus and document management. Event-driven where systems allow, with batch and polling fallback for legacy. One common behaviour handles auth, field mapping, retries, sync and observability.
When you keep your existing systems, they stay the source of truth and KrimOS writes back only on validated channels. When KrimOS is the stack, the same gate holds and the source of truth is its own immutable record.
Two paths in
- Event + stream: webhooks and streaming for systems that emit changes in real time.
- Batch + poll: scheduled extracts and polling for legacy systems that don’t.
- Validated write-back: nothing returns to your systems until it has cleared the gate.
One architecture, every geography
The same runtime, sovereign in every market.
KrimOS runs the same architecture in every market it serves. Only Krim-Fabric’s rule set changes. Each region runs self-contained behind its own walls, the local law already loaded, so one runtime, audit trail and governance carry from market to market unchanged.
Common questions
What engineers ask first.
Does KrimOS replace our core banking, LOS or CRM?
Your choice. KrimOS is a complete lending stack of its own — origination, underwriting, servicing and collections — so you can run the operation on it end to end. Or keep the systems you have: KrimOS layers on top, reading from them and writing back only on validated channels, so you can start with nothing to tear out and run as much on KrimOS as you choose.
How does KrimOS connect to our existing stack?
Through an integration fabric of 40+ connectors across core banking, origination and servicing, CRM, telephony, messaging, payments, credit bureaus and document management. It is event-driven where systems allow, with batch and polling fallback for legacy.
Does the same architecture run in every market?
Yes. One architecture runs in every market; only Krim-Fabric’s per-jurisdiction rule set changes. Each region runs sovereign within its own perimeter, with the jurisdiction’s law already encoded.
See it slot into your architecture.
A two-week technical deep-dive covers architecture, security and integration, then a scoped pilot follows on ring-fenced data.