Symbiosis Knowledge ingests docs, wikis, tickets, drives, chat, code, and databases, then serves permission-aware hybrid search and citation-grounded answers, or an honest not-found. It continuously evaluates faithfulness, finds content gaps, supports expert curation, and exposes APIs that ground Voice, Flow, Learn, Agent Cloud, and Ask Symbiosis. An answer without a citation is just a confident rumor.
Enterprise AI search usually fails in one of three ways: it invents, it leaks, or it goes stale. Knowledge is built against those: every claim links to a passage, retrieval enforces source ACLs per user, and freshness plus eval harnesses catch drift before users lose trust.
RAG is a component. Knowledge is the system around it: connectors, chunking, hybrid lexical and vector retrieval with reranking, ontology, expert curation, multi-collection tenancy, grounding APIs for other products, and a content-gap report that turns unanswered questions into a documentation roadmap.
Ask Symbiosis on this site is Knowledge on our own corpus: permission-aware retrieval, cited answers, honest not-founds. In product, the same plane powers employee assistants, customer help, agent grounding, voice answers, and Flow decision steps.
Ask Symbiosis- the assistant floating at the bottom-left of every page - is Knowledge running on this site's own content: permission-aware retrieval, cited answers, honest "not found"s. Ask it something and watch the citations.
Connector fleet & continuous ingestion
Wikis, drives, PDFs, tickets, chat, email archives, code repos, databases, and APIs - chunked, enriched, embedded, and re-indexed on freshness signals.
Permission-aware retrieval (ACL mirroring)
Index ACLs with content; enforce at query time per requesting user/agent. Search and answers never become a permissions bypass.
Hybrid search + reranking
Lexical + vector retrieval with rerankers - strong on IDs, acronyms, policy codes, and natural language alike. Per-source authority weighting.
Cited answers & honest abstention
Every claim linked to passage(s); unsupported questions get 'not found' or escalate - never improvised policy.
Multi-collection & tenancy
Collections per team/entity/product with shared vs private corpora, residency fences, and cross-collection query policies.
Ontology, taxonomy & entities
Living knowledge graph overlays: products, policies, people, systems - so retrieval understands your vocabulary, not just embeddings.
Knowledge capture & distillation
Capture from calls, chats, tickets, and retiring experts; structure into durable articles with ownership and review cadences.
Expert-in-the-loop curation
Review queues for new chunks, conflicting sources, low-confidence answers, and proposed article drafts - with override trails.
Freshness scoring & stale quarantine
Detect stale, superseded, or conflicting docs; quarantine from answers until an owner refreshes or retires them.
Evaluation harness (faithfulness gates)
Continuous suites for faithfulness, citation support, abstention quality, hit-rate@k, MRR - regressions block deploys.
Content-gap & query analytics
What people ask that docs can't answer becomes your roadmap; zero-result and low-faithfulness clusters ranked by volume.
Grounding API for agents & apps
Retrieve-and-cite endpoints for Agent Cloud agents, Flow Knowledge steps, Voice dialogue, Learn tutors, and custom apps.
Enterprise search surface
One search box over the systems people already live in - permission-filtered results with authority and freshness signals.
Assistant & channel surfaces
Embedded chat, Slack/Teams bots, portal widgets, and Ask-style assistants with the same citation contract.
Document intelligence handoff
Layout-aware extraction for scans/PDFs into the corpus; optional validation hooks before indexing.
Conflict & multi-source resolution
When sources disagree, surface the conflict with citations instead of picking a fluent lie - escalate to owners.
PII / secret redaction pipeline
Detect and redact or segregate sensitive spans before embedding; residency and retention policies per collection.
Model-router aware generation
Answer generation via Agent Cloud model routing when present - cost/latency/residency policies apply; retrieval stays the source of truth.
Air-gap & private deployment
Self-hosted / VPC options for regulated estates; exportable audit of queries, retrievals, and answers.
Benchmark & evidence packs
Reproducible eval reports for security/legal review - methodology, golden sets, and scores you can re-run.
INGEST FABRIC
Connector fleet, chunking strategies, enrichment, embeddings, freshness-driven re-index.
PERMISSION PLANE
ACL mirroring, identity sync, query-time enforcement for users and agents.
RETRIEVAL ENGINE
Hybrid lexical+vector search, reranking, authority weighting, multi-collection query.
ANSWER PLANE
Cited generation, abstention, conflict surfacing, channel assistants (web/Slack/Teams).
KNOWLEDGE GRAPH & TAXONOMY
Entities, ontologies, synonyms, and vocabulary that match how your org actually talks.
CAPTURE & CURATION
Tribal-knowledge capture, expert review queues, freshness quarantine, article lifecycle.
EVAL & GAP OS
Faithfulness regression suites, content-gap analytics, deploy gates, evidence packs.
GROUNDING API
SDK/API for Agent Cloud, Flow, Voice, Learn, and third-party apps - retrieve, cite, abstain.
Investment is bespoke to estate size, compliance needs, and deployment shape - book a demo and we'll map the right fit.
STARTER
- → Core connectors (docs/wiki/drive)
- → Hybrid search
- → Cited answers + abstention
- → ACL-aware retrieval
- → Basic eval suite
- → Embedded assistant widget
BUSINESS
- → Everything in Starter
- → Full connector fleet + tickets/chat/code
- → Content-gap analytics
- → Curation queues & freshness quarantine
- → Slack/Teams bots
- → Grounding API for agents/Flow/Voice
- → Continuous eval gates
ENTERPRISE
- → Everything in Business
- → Knowledge graph & taxonomy studio
- → Multi-collection tenancy + residency
- → PII/secret redaction pipelines
- → Conflict resolution workflows
- → Benchmark evidence packs
- → Private / air-gapped deployment
Knowledge, answered
Can it answer from data that users aren't allowed to see?
No - that's the point. Retrieval enforces source permissions per requesting user (and per agent identity), so the assistant can't become a permissions bypass.
How is Knowledge different from 'just RAG'?
RAG is the retrieve-then-generate pattern. Knowledge is the OS around it: connectors, ACL mirroring, hybrid search, curation, freshness, eval gates, gap analytics, and grounding APIs for the rest of your stack.
What stops hallucinations?
Grounding plus evaluation plus honesty affordances. Answers must cite retrieved sources; faithfulness is measured continuously; the system is designed to say 'not found' rather than improvise. Eval regressions block deploys.
Does it replace our wiki / SharePoint / Confluence?
No - it indexes and serves them (and more) under permissions. Authors keep writing where they write; Knowledge makes that content findable, citable, and safe to answer from.
How does it power other Symbiosis products?
Voice grounds dialogue in Knowledge; Flow can require cited policy before a decision; Learn tutors answer from curriculum; Agent Cloud agents call the Grounding API instead of inventing facts.
What is a content-gap report?
A ranked list of questions users ask that retrieval can't support - your documentation roadmap ordered by real demand, not by who shouted in the last meeting.
See it against your reality.
A 30-minute walkthrough with an engineer who builds it - your use case, not a script.