Connect
Ingest documents, databases, object stores, service logs, MCP resources and engineering systems through governed, module-gated connectors.
Platform
Turn Fragmented Enterprise Data into Governed, Traceable Intelligence
kiLM connects documents, engineering systems, operational data, knowledge graphs and local AI models inside a dedicated single-tenant environment — so teams can investigate, validate, simulate and act with evidence behind every decision.
From Connected Data to Approved Action
One governed lifecycle
Every capability works together in one pipeline — not as isolated features. Data moves left to right; a governance rail holds underneath at every step.
Normalize
Parse, chunk, classify sensitivity and resolve entities into consistent, versioned records with lineage preserved.
Link & Reason
Build the trace graph and knowledge graph — connecting requirements, designs, parts, tests, suppliers and policy across domains.
Answer & Analyze
Retrieve with vector, keyword, graph and visual signals; run change-impact, what-if and validation with citations behind every result.
Advise, Review and Approve
Propose cross-system actions — tickets, record updates, workflows — each policy-checked and held for human approval before execution.
Improve
Capture feedback, evaluate answers and models, and promote or roll back improvements under governance.
Single tenant
Source & role controls
Provenance
Policy
Human approval
Audit
Six Pillars, One Governed Platform
Capabilities
Every capability is labelled on two dimensions — availability and any dependency:
Availability
Available in Core
Available as Option
Available as Preview
Planned
Dependency
No additional connector
Connector required
Customer data or configuration required
Live validation required
Enterprise Data & Connector Fabric
Foundational- Multimodal ingestion of documents, office files, images, geospatial and structured records.
- Enterprise connectors with revisions, deltas and attachment lineage.
- Federated queries and schema discovery across systems.
- Per-connector quality and health signals.
Trace Graph & Engineering Digital Thread
Optional module- Requirements-to-design-to-test traceability across connected domains.
- Revision and effectivity awareness with trace coverage.
- Change-impact analysis and read-only what-if simulation.
- Audit packs assembled from cited source records.
Grounded & Neurosymbolic Intelligence
Foundational- Vector, keyword, graph and visual retrieval blended for recall.
- Ontology bindings and rule checks (SHACL / OPA) over governed data.
- Citations, conflicting-evidence detection and bounded refusal when evidence is insufficient.
- Grounded answers tied to source records — evidence, not assertions.
Governed Agents & Cross-System Actions
Optional module- Agent Studio with governed templates, typed inputs and outputs.
- Tool allowlists, budgets and full run traces.
- Approval-gated write-backs to Jira, ServiceNow, Slack and MCP — Preview: recorded execution today; live adapters not yet shipped.
- Every action proposed, policy-checked, approved and audited.
Model Lifecycle & Domain AI
Optional module- Local model registry with task-aware selection.
- Fine-tuning: SFT, LoRA / QLoRA, DPO / RFT and evaluation.
- Domain SLM distillation — Preview: contracts, safeguards and deterministic cores are shipped; live training and serving are being wired.
- Controlled promotion and rollback; concurrent multi-model serving at scale is on the roadmap.
Sovereign Enterprise Operations
Foundational- On-premises, managed single-tenant and air-gap deployment (air-gap when images, dependencies and model weights are pre-staged).
- Identity, sensitivity, retention and erasure controls.
- SBOM, monitoring and capacity forecasting.
- Dedicated single-tenant isolation — your data and models stay in your environment.
What Makes kiLM Different
Differentiation
Evidence, not merely answers
Citations, source records, trace paths and policy verdicts back every result.
Engineering depth
CAD, PLM, BOM, requirements, simulation, test, quality and field records — not just documents.
One governed lifecycle
Ingest, extract, link, retrieve, reason, act, evaluate and improve — in a single controlled pipeline.
Bounded agency
Actions are proposed, policy-checked, approved and audited — never unrestricted autonomy.
Sovereign by design
Dedicated single-tenant deployment with local-model and air-gap options.
Built to Show Its Work
See it in context
These are actual product surfaces. Each capability exposes its evidence, severity and approval trail — not a black box.
Trace graph with evidence drill-down
Follow any requirement to its design, test and field records, and open the source behind each edge.
Change-impact graph with severity & confidence
See what a change touches across domains, ranked by severity and confidence, before you approve it.
Agent Studio with approval & execution trace
Watch a governed workflow propose an action, pass policy checks, and wait for human approval before it runs.
Chat with visuals & cited sources
Ask in plain language and get a grounded, cited answer alongside the diagrams, scans and images kiLM matched by their visual content.
What-if change simulation
Model a proposed change against your governed data, compare alternatives, and see the regression checklist it would trigger — before you commit.
Compliance & obligation tracking
Watch a corpus against its obligations, file findings, and track each remediation to closure — every item linked back to its source clause.
Explore the Platform
Explore
A guided tour of everything under the Platform menu — skim each area here, then open the full page whenever you want the detail.
Connect kiLM to the Data Your Teams Already Use
The connectors, protocols, and cloud services kiLM ingests — each module-gated.
Read the full page →Built for Governed Data from Day One
Sensitivity classes, field-level encryption, cascade erasure, and a full audit trail.
Read the full page →Runs Where You Run
A self-contained stack you host yourself, from a CPU-only trial to a GPU production cluster.
Read the full page →Run kiLM the Way Your Security Model Demands
The same image set in every topology — fully isolated, cloud, or hybrid.
Read the full page →Permissive by Default, Copyleft Only by Explicit Choice
Predictable licensing — the default stack uses only permissive components.
Read the full page →Sign in with the Identity You Already Run
Federates to your identity provider — OpenID Connect, SAML 2.0, SCIM 2.0, and LDAP / Active Directory.
Read the full page →Any LLM. Your Choice, per Task.
Model-agnostic by design — run open-source, open-weight, frontier, or paid models.
Read the full page →
Your Data and Derived Models Remain Yours
Ownership & IP
As between you and POLYCRACY (kiLM's maker), POLYCRACY does not claim ownership of your inputs or your customer-specific outputs — the governed corpus, the intelligence derived from it, and models fine-tuned or distilled on your data. POLYCRACY's own intellectual property is the kiLM platform itself. This allocation is subject to the licenses of any third-party or open-weight base models, third-party source-document rights, your customer agreement, and applicable law.
- As between the parties, your data and everything derived from it — extractions, entities, graphs, embeddings, indexes and reports — is treated as yours.
- Adapters, preference-tuned models and distilled domain SLMs trained on your data are yours as between the parties, subject to the licenses of any open-weight base models and any third-party source rights.
- POLYCRACY claims intellectual property only in the kiLM platform — its software and tooling. Base models kiLM ships remain governed by their own upstream licenses.
- Your inputs and outputs are not used to train shared or foundation models; they stay inside your dedicated, single-tenant deployment.
See kiLM on Your Own Data
Tell us your stack and the decision you need to support. We'll scope a governed starting point — on infrastructure you choose.