Ask your knowledge, like a person.
Ingest company documents as searchable knowledge, and let a team of expert agents — each with a name and a role — answer with sources and citations. Aggregation, comparison, policy rulings and multi-turn dialogue: every figure is verified by deterministic computation, for a zero-hallucination knowledge system.
Locus KMS is an enterprise knowledge management system that turns company documents into searchable knowledge and answers through a team of expert agents with names and roles — always with sources and citations. It goes far beyond search: aggregation, comparison, exhaustive extraction, policy rulings, multi-step calculation and multi-turn dialogue, running the whole loop of ingest → search → answer → knowledge enrichment in one system.
Policies, prices, procedures and per-client exceptions live in one person's head and across scattered documents. When they're away, even familiar questions have to be re-checked.
A list of documents can't answer questions that need a definitive value, like a price or a policy. Users end up opening, comparing and interpreting documents all over again.
Which questions went unanswered? Which areas have no documents at all? Knowledge is not something you accumulate — it's something you operate.
Upload PDF, Word, Excel, PowerPoint and HWP documents, images or web pages, and they are parsed, block-extracted, classified and embedded into knowledge. Track each ingest stage in real time in the operations console; failures show the exact failing step and can be reprocessed from any selected stage in one click.
"Gyuri, summarize the annual-leave policy." Agents with names, roles and assigned knowledge scopes answer within their own domain, attaching source documents and citations. A coordinator agent takes your request, delegates to the right specialist and consolidates the results.
Filtered aggregation, multi-step calculation, set relations, document comparison, exhaustive extraction, version lineage and policy rulings — the engine recognizes the question type and routes it to the right decision path. Every figure is computed deterministically with read-only SQL — zero hallucination. Answers no top-k retrieval sample could produce.
"Top 5 accounts by quote value" → "only the largest one." Keep refining a result within the session. Elliptical follow-ups inherit the context, while new topics start clean without contamination.
Manage by repository, category, document type and status; sort by search hits to see which documents are actually used; view and edit at block level, include or exclude blocks from search, and reprocess.
Structure prices, terms and policies under canonical keys, then layer scope overrides by branch, client and period. A resolution simulator lets you pre-verify which value answers under which conditions.
Unanswered queries are diagnosed as knowledge gaps; agent-written drafts flow through write → review → approve/reject, and only approved drafts become knowledge. The knowledge roadmap tab shows the enrichment plan alongside.
Start from the roster, configure persona, knowledge scope, tools and channels, and validate with test chat. Invocation logs and audit logs, all on one screen.
Three steps
Upload PDF, Word, Excel, PowerPoint, HWP, images or web pages — they are parsed, block-extracted, classified and embedded into searchable knowledge. Watch progress live on the pipeline screen.
Direct an agent with a name and a role, and it answers within its knowledge scope with source documents and citations. Definitive values like prices and policies are resolved by the master-data layer, scoped down to branch and client.
Unanswered queries are collected automatically and diagnosed as knowledge gaps; agent-written drafts pass approval or rejection before becoming knowledge.
From consultation to tailored deployment — Timbel works with you.