What Syncratic is
A Knowledge Assurance System
Syncratic represents enterprise data as a network of relationships, not isolated documents or fragments. Entities, roles, dependencies, and changes over time are interwoven into a living knowledge structure that can be reasoned over, not just retrieved.
Platform surfaces
One workspace, every knowledge motion
Each surface exists to make one kind of knowledge work inspectable: discovery, reasoning, relationship review, change review, and operations.
Search / Ask
A unified discovery-to-reasoning surface. Search finds evidence across accessible scopes; Ask reasons over it with citations, provenance, and memory threads.
Insights
Graph-derived relationship narratives: live streams, CRM signals, and document deltas clustered into readable, subject-driven insights.
Explorer
Document inspection with metadata, graph context, and change history: see the evidence behind every answer at the document level.
Document Delta Engine
Version lineage and deterministic deltas. When a contract, policy, or filing changes, see what moved, when, and why it matters.
AI Gateway
A designed capability: applications call Syncratic Ask through a governed API, with tenant policy, privacy checks, spend limits, and audit evidence on every call.
Connectors
Governed source synchronization with schedules, sync state, per-file visibility, and administrator-controlled configuration.
Privacy Engine
Runtime context governance that masks, reduces, or withholds sensitive evidence before it reaches a model, without mutating canonical artifacts.
Audit & Admin
Boundary diagnostics, event inspection, and operational traceability. RBAC-controlled administration of users, roles, licensing, and models.
Semantic search and retrieval
Beyond file titles, into understanding
Most enterprise search matches words. Syncratic reasons over relationship structure. Every query is decomposed into its semantic parts, so retrieval follows meaning across the enterprise corpus, at scale, instead of stopping at keywords in a filename.
The result is search that understands context: direct questions, reverse questions, and follow-ups all retrieve the right evidence, because the system models how facts connect rather than how documents are titled.
- Hybrid retrieval: lexical, vector, and graph evidence, ranked together
- Document-family-aware enrichment for contracts, filings, cases, and datasets
- Direct, reverse, and follow-up questions all retrieve the right evidence
How a query is understood
Every query is decomposed into its semantic parts:
Knowledge association across subjects, objects, and verbs: the goal is to understand relationship structure algorithmically, not to match hardcoded words.
Governed connectors
Unstructured sources in. Governed knowledge out.
Syncratic turns unstructured and semi-structured enterprise documents into governed, searchable, explainable, and evolving organizational intelligence. Enterprise systems (SharePoint libraries, Google Drive, OneDrive, Gmail, Salesforce CRM) are treated as governed knowledge sources, not one-off imports.
- ACL-aware SharePoint ingestion: source access signals travel with content, so downstream visibility respects what each user is allowed to see.
- CRM federation: Salesforce accounts, opportunities, contracts, and cases become connected knowledge, not just a sales database.
- File-share estates: deployment-mounted NFS, SMB, and CIFS shares ingest with governed sync, provenance, and downstream boundaries.
- A shared lifecycle: every connector runs the same governed flow of source-appropriate authorization, scoped synchronization, staged ingest, and continuous refresh.
Sources
- Google Drive
- Gmail
- Salesforce
- OneDrive
- SharePoint Online
- NFS / SMB / CIFS
Google Drive, Gmail, and Salesforce are live in production. OneDrive and SharePoint are supported as scoped deployment tracks when enabled.
Connector lifecycle
- Source-appropriate authorization, per provider
- Scoped synchronization with schedules you control
- Staged ingest with per-file state and retries
- Sync observability and stale-run recovery
Governance
RBAC and data boundaries, enforced: not suggested
Access control in Syncratic is enforced at query time across every store: documents, graph, CRM, and insights. Tenant and user scopes define what can be retrieved; object-level access defines what can be seen; role-based access governs who administers what.
- Identity through Keycloak/OIDC with role and identity context on every request
- Tenant/user scope and object-level access enforced across all data stores
- RBAC-controlled administrative surfaces: users, roles, licensing, models, privacy
- Privacy Engine gates sensitive evidence at the model context: mask, reduce, suppress, or deny
- Boundary diagnostics in Audit: see exactly when and why a boundary was applied
Principle
Every answer should be defensible. If you cannot explain why the system answered this way, it is not production-ready.
What boundaries mean here
A user searching in tenant scope sees only what object-level access permits: private material stays private, and cross-user leakage is prevented by boundary controls, not policy documents. The Privacy Engine protects model context without corrupting the knowledge base; canonical artifacts are never destructively redacted.
Deployment posture
Sovereign by architecture, model-agnostic by design
Sovereign deployment
Offline-safe licensing and entitlement enforcement keep the deployment self-contained while premium capabilities stay governed and auditable. Run air-gapped, own your infrastructure, and keep operational control, with admin bootstrap, SMTP-backed invitations, license update, and readiness evidence included.
Model-agnostic by design
The model layer is intentionally replaceable. SLMs and LLMs are replaceable execution components: model roles can be bound to private, deployment-local LLMs or external public endpoints, with a trust zone available to Privacy Engine policy. You choose the deployment posture per role rather than being locked to one model location or provider. The durable value is the system above the model layer.
What's included
Production capabilities, not roadmap promises
- Evidence-based answers with source-aligned citations and provenance estimates
- Tenant and user data-boundary controls enforced at query time
- Role-based access for administrative surfaces, via Keycloak/OIDC identity
- Human-guided knowledge evolution: feedback becomes durable improvement
- Continuous, governed connector sync across approved enterprise sources
- API-first observability with audit trails and boundary diagnostics
- Sovereign deployment: offline-safe licensing and air-gapped operation
- Model-agnostic by design: bind model roles to private or public endpoints
Questions
Frequently asked
Which sources does Syncratic connect to?
Google Drive, Gmail, and Salesforce are live in production. OneDrive and SharePoint Online 365 are supported as scoped deployment tracks when enabled, and deployment-mounted NFS/SMB/CIFS shares are supported for file-share estates. Every source runs through the same governed connector lifecycle.
Where does our data live?
In your deployment. Syncratic runs as a tenant-controlled platform with Helm/K3s production posture; the workspace, its stores, and its licensing remain self-contained. There is no dependency on a vendor SaaS to operate.
Can we run Syncratic air-gapped?
Yes. Offline-safe licensing and entitlement enforcement are built in, so premium capabilities stay governed and auditable without external calls. Model roles can be bound to deployment-local LLMs for fully sovereign operation.
Do answers cite their sources?
Yes. Generation is citation-constrained: answers are built strictly on retrieved evidence with provenance estimates, and you can inspect the prompt evidence window to see exactly which evidence was admitted or excluded.
What happens when a source document changes?
The Document Delta Engine captures version lineage and deterministic deltas. Material changes surface in Insights and Explorer, including contract and procurement deltas and Salesforce contract-value evolution, so review focuses on what actually moved.
Are we locked into one model provider?
No. The model layer is intentionally replaceable. Model roles (basic reasoning, deep reasoning, classification, evaluation) can be bound to private, deployment-local LLMs or external public endpoints, so you choose the deployment posture per role instead of being locked to one provider.
Governed knowledge assurance
See Syncratic on your own data.
A 30-day pilot: scoped ingest baseline in week one, adoption with citation review in week two, feedback-driven quality iteration in week three, and a value review in week four.