Public-source analysis,
linked to the evidence behind it.
Cosint builds Sentra, an analysis workspace for organizing public information into entities, timelines, graphs, and reviewable reports.
Attributes 5
Evidence chain designed · 4 entries
Copilot · grounded
Entity A-1042 appears on Filing F-318 alongside Sample Org B1. A related ownership record names Sample Org A2. Confidence: probable3.
Resolve who, where, and what across noisy sources.
Typed entities — person, org, location, event, vessel, source signal, registry, infrastructure — with deduplication, alias capture, and source-rated confidence.
Relationships in a typed graph, on a map, on a timeline — synced.
Force-directed graph, geospatial workspace, and multi-lane timeline share one selection. Cross-pane pulse keeps three views aligned.
Source-linked outputs with a designed evidence chain.
Every change captures who, when, and what changed. Source reliability (A–F) and information credibility (1–6) follow each claim through to export.
Reviewable briefs with footnoted, source-linked findings.
Drag-and-drop report builder. Calibrated confidence vocabulary, review markings, and a metadata trail on every export.
Six stages from public sources to a reviewable report.
Sentra is built around how analysts actually work. Each stage produces traceable artifacts the next stage can act on.
Search
Hybrid BM25 + vector search across ingested public sources and uploaded documents.
Extract
Named-entity extraction with type, span, and source pointer captured as evidence.
Resolve
Deduplicate, alias-match, merge, and rate. Analyst-in-the-loop before any merge sticks.
Analyze
Graph traversal, centrality, association inference, supply-chain mapping.
Map
Plot to a typed geospatial canvas. Tracks, review areas, geofences, clustering.
Report
Draft source-linked briefs with footnoted claims, calibrated confidence, and review metadata.
Inside the workspace.
Four core surfaces. Each one is a real Sentra view — graph and map analysis, source-linked search with retrieval-augmented generation, entity extraction with relationship inference, and source-linked report generation.
Graph, map, and timeline share one selection.
Pick an entity in any pane — the other two pulse to it. ADR-004 "Pulse" keeps three views in sync so analysts can hold a single thread across relationships, geography, and time without rebuilding context.
- Typed graph with eight entity classes and confirmed/suspected edges
- Geospatial canvas with review areas and tracks
- Multi-lane timeline with point and span events
- Source reliability (A–F) and information credibility (1–6) visible everywhere
Hybrid keyword + vector retrieval. Source-linked every time.
Keyword and vector retrieval run in parallel against a full-text index and a workspace-scoped vector index. Results carry source reliability ratings, entity highlights, and inline citations into the Copilot.
- Hybrid scoring with workspace-scoped corpora
- Document ingest with chunking, OCR fallback, and provenance
- Saved queries with scheduled alerts
- RAG context never leaves the analyst's workspace
From paragraph to typed graph, with the source span attached.
Extracted entities carry the exact passage they came from, the model that proposed them, and the analyst who approved them. Relationship inference proposes edges; the analyst confirms before they are saved.
- Eight entity types — person, org, location, event, vessel, source signal, registry, infrastructure
- Inline source spans preserved for every claim
- Proposed edges marked suspected until reviewed
- Graph-backed analysis for centrality, traversal, supply-chain mapping (target architecture)
It is probable PROBABLE · 55–80% that Sample Org A and Sample Org B share a beneficial-ownership structure linked to Entity A-10421, based on a public registry filing dated 2024-08-142.
The judgement rests on two corroborating A/1 public registry sources3 and one B/2 trade-press account4. A single D/4 open-web indicator is excluded from the central claim.
Source-linked briefs with calibrated confidence and a designed evidence trail.
Build the brief from the workspace. Every claim carries its source pointer and its place in the evidence chain. Confidence uses a calibrated vocabulary so reviewers can challenge a number, not a vibe.
- Template with key judgement and confidence basis
- Footnoted sources resolve to live evidence packets
- Review markings persist into PDF and HTML export
- Audit metadata on every exported artifact
Built for review.
Sentra is designed for public-information analysis by trained analysts. The product surfaces are shaped so AI assists, humans decide, and every output can be traced back to where it came from.
Public-source focus
Designed around public information, with workspace-scoped storage and explicit out-of-scope guardrails in the use policy.
Analyst-in-the-loop
AI proposes, the analyst decides. No silent finalization of merges, claims, briefs, or alerts.
Source-linked outputs
Every claim resolves to the source span it came from. Reliability (A–F) and credibility (1–6) follow the claim through to export.
Audit & evidence trail
Designed evidence chain, append-only audit log on operator actions, review markings on every export.
Concrete workloads. Provider-neutral.
Sentra needs cloud infrastructure for a specific set of workloads — web runtime, managed databases, graph store, vector index, geospatial services, AI inference, and audit storage. See the printable brief.
| # | Workload | Why Sentra needs it | Category |
|---|---|---|---|
| 01 | Web & API runtime | FastAPI backend + React frontend with horizontal scaling and per-workspace deployment slots — prototype direction. | compute |
| 02 | Postgres + PostGIS | Canonical store for entities, investigations, geospatial queries, audit log; PITR and encrypted backups. | data · geo |
| 03 | Graph search | Typed graph for relationships, traversal, centrality, supply-chain mapping — target architecture. | graph |
| 04 | Vector search | Embeddings for hybrid search, document RAG, and Copilot retrieval — workspace-scoped target architecture. | retrieval |
| 05 | Object storage | Workspace-scoped storage for source documents, capture artifacts, exported reports. | data |
| 06 | AI inference | Bounded inference for extraction, summarization, RAG grounding, and ongoing evaluation. | gpu · inference |
| 07 | Observability & audit | Structured logs, traces, append-only audit storage — scoped to workspace boundaries. | observability |
A focused software company.
Cosint builds Sentra, a public-source analysis workspace for source-linked research, review, and reporting.
General & design partners
For product conversations, pilots, and engineering inquiries.
hello@cosint.ioSecurity disclosure
Responsible disclosure of vulnerabilities or platform concerns.
security@cosint.io