Geospatial Intelligence Fusion

The world,
queryable.

Request access →See it live →
Maritime · live vessel positionsEnergy · grid load & cross-border flowSanctions · designations & watchlistsNews · open-source global newsAviation · live flight positionsResearch · scientific literatureWeather · forecast & sea stateInvestigations · entities & documentsMaritime · live vessel positionsEnergy · grid load & cross-border flowSanctions · designations & watchlistsNews · open-source global newsAviation · live flight positionsResearch · scientific literatureWeather · forecast & sea stateInvestigations · entities & documents
▌  01 / THESIS

Decisions about the physical world are still made from PDFs, screenshots, and stale slides. insigz is the layer between live infrastructure and the people who have to act on it — a single canonical model where every observation, every entity, every event can be queried, traced, and reasoned about.

FUSED FROMMARITIME · ENERGY · SANCTIONS · NEWS · RESEARCH
Maritime
Vessel positions, port states & identity
Aviation
Live flight positions & airport reference
Sanctions
Designations & consolidated watchlists
Energy
Grid mix, cross-border flows & prices
Weather
Forecast, wind & sea state
News
Open-source global news
Research
Scientific literature & outbreak signals
Investigations
Entities, ownership & documents
◐ PREVIEW · ONTOLOGY AGENT Onboard a new source by chatting with an agent that profiles the feed, drafts the connector and field→Observation mapping, fits it to your existing model, and dry-runs it — a human commits every schema change. Available today as a guided demo. SEE THE PREVIEW →
02b / WHO USES IT

Six ways in, one platform.

/01
Maritime intelligence
Marine insurers, P&I clubs, brokers, port authorities. Vessel identity, sanctions exposure, AIS history — one query, one minute.
LEARN →
/02
Energy market intelligence
Utilities, traders, regulators, infrastructure operators. Grid mix, cross-border flows, prices & weather, fused live for the question you're asking now.
LEARN →
/03
Sanctions & export-control
Banks, corporates, audit firms, NGOs. Beneficial-ownership traces, cargo paths, audit-grade chains — every claim cited.
LEARN →
/04
Investigative journalism & OSINT
Newsrooms, think tanks, OSINT collectives. Per-tenant means investigations stay confidential; the evidence chain stays defensible.
LEARN →
/05
Academic & policy education
Universities, defense academies, think-tank war games. Strategy-Day-ready scenarios; the canonical example is the Baltic cable-cut.
LEARN →
/06
Research & evidence
Pharma, public health, due diligence. Turn literature and reports into a source-anchored evidence base — findings kept distinct from mentions.
LEARN →
● WHO, EXACTLY

We don't name our customers. They're mid-to-large private and listed companies — and a few public institutions.

The organizations that get the most out of insigz work in domains where being named on a vendor's website is a liability, not a logo win. So we keep the client list private by default — and treat that discretion as part of the product, not a gap in our marketing. Under NDA, we're happy to share relevant references directly.

WHY WE STAY QUIET →
02 / DATA MODEL

One canonical chain from raw source to delivered insight.

Every piece of intelligence in insigz traces back along the same six-step chain. No bespoke pipelines, no silo'd buckets — sources flow into observations, observations bind to entities, entities accumulate events, events resolve into cases, cases publish as reports.

Wireframe diagram showing six nodes connected left to right above a perspective dotted floor
/01
Source
Maritime, energy, sanctions, news, weather & OSINT feeds, partner APIs.
/02
Observation
A timestamped, geolocated atomic fact. Immutable, signed, traceable.
/03
Entity
A vessel, a cable, a substation, a sanctioned party — the noun the world is made of.
/04
Event
A change. AIS-gap, grid-frequency-excursion, designation, anomalous routing.
/05
Case
A bound collection of events that mean something together. The unit of analysis.
/06
Report
The artifact that leaves the platform. Signed, cited, time-stamped.
Baltic Sea visualized as data — coastlines as thread-lines, submarine cables as light, vessels as drifting particles
59.4370° N · 24.7536° E
GULF OF FINLAND · SECTOR 7
● LIVE
AIS live
GRID live
SANCTIONS live
03 / Live world layer

This is not a slide.
It's the world, breathing.

Electrical grid corridors, vessel positions, port states, market signals, sanctions designations, open-source news — all live, all geolocated, all queryable through the same model.

The seed for any session, scenario, or report is a single frozen moment of this world.

MaritimeEnergySanctionsNewsWeatherAviationOSINT
04 / INTELLIGENCE

Three AI roles. Analyst-grade oversight on every output.

insigz's intelligence layer is built around three specialised agents — none of which publish autonomously. Every suggestion shows its reasoning, cites its source observations, and waits for a human in the loop. AI assists, never decides.

Wireframe diagram showing the three AI agent roles connected to a central session node
/04.1 ADJUDICATOR ◐ PREVIEW

Draft consequences from actions.

Given a proposed action and the relevant slice of world state, the Adjudicator produces 2–3 plausible consequence variants — each anchored in historical analogs, each with a confidence range, each with full reasoning visible.

Human-in-loop: consequences queue for analyst review. Nothing publishes to downstream consumers until approved or edited.
/04.2 ANALYST ● LIVE

Answer questions, grounded in the data.

The analyst chat runs Claude over the canonical model with tool access — it queries observations, entities, events and the document corpus to answer in plain language. Every claim cites the source observation it rests on.

Grounded: answers must cite real observations; uncited or dangling claims are rejected and regenerated. No ungrounded prose.
/04.3 AFTER-ACTION ◐ PREVIEW

Draft the report; the analyst signs.

At case close, the After-Action analyst drafts the full report: timeline, key decision points, base-rate benchmarks, discussion prompts. Every claim carries a citation back to a source observation.

Human-in-loop: the draft is editable; faculty or analyst signs the final artifact. Their name is on every outcome.
THE AUTONOMY LAYER · LIVE
It watches for you. It never acts on its own.
Standing Watches run grounded agents on the live feed and stage Proposals; a human Accepts, Modifies or Rejects. Trust Evals prove an agent before it ships, and Analyst Studio builds new grounded agents without code. Autonomy is bounded and reversible — only internal, reversible outcomes ever auto-apply (doctrine Q10). READ MORE →
05 / SESSIONS

One world. Many viewpoints. Enforced asymmetry.

Sessions are an extension on top of the core platform — used by universities, defense academies, corporate scenario teams, and policy institutes for structured tabletop exercises. Each cell sees a different slice of the same world; information asymmetry is enforced by the data model, not by trust. The canonical example is the Baltic cable-cut Strategy Day, documented in depth here →.

Wireframe diagram of four role figures arranged around a central wireframe Baltic outline
NATO / EU CELLCoordinates political-military response. Sees national intelligence summaries, operator reports, sanctions levers.
HYBRID THREAT CELLPursues confidential strategic objectives. Sees the threat plan, shadow-fleet positions, deniability budgets.
CIVIL / MEDIA CELLRepresents public discourse and civilian impact. Sees sentiment indicators, infrastructure status, news ecosystem.
FACULTY / ADJUDICATOROmniscient. Approves consequences, fires injects, advances phases. The final word, every time.
06 / AFTER-ACTION

Every case, every session, every report — closes with an artifact.

The After-Action analyst replays the full timeline, identifies key decision points, benchmarks against historical base rates, and ships a printed report — the artifact that leaves the room.

Wireframe pipeline showing session timeline flowing through an analyst sphere into a report document
/06.1

Session timeline

Every action, every consequence, every inject, every approval — recorded with timestamps and reasoning chains. The basis of the audit.

/06.2

Analyst pass

The agent identifies the 5–8 most consequential decisions, generates a base-rate benchmark for each, and drafts discussion prompts for the next session.

/06.3

Signed artifact

A printed report — generative cover, editorial inside — delivered as a signed PDF. Faculty edits; faculty signs; faculty owns.

REPORT

A document worth keeping.

Every report's cover is a generative data sculpture seeded from the session's UUID — visually unique, anchored to the case it documents. Inside: dense editorial typography, base-rate citations, full reasoning logs in the appendix for motivated readers.

Faculty and analysts cite it. Students keep it. Procurement files it.

Signed
EVERY SESSION
From session close to a signed PDF in your inbox — the delivery window is agreed up front.
07 / SYSTEM

How insigz works.

Four layers, in series. Live data fuses upward into a canonical model; the agent triad reasons over it; sessions and reports leave the platform as signed artifacts. No proprietary data formats, no black-box pipelines.

/01
Fusion engine
INGEST · NORMALIZE · GEOLOCATE
  • Live sources polled or streamed — maritime, energy, sanctions, news, research
  • Dedupe by entity + temporal window
  • WGS84 geometry on every observation
  • Provenance signed, traceable, immutable
/02
Canonical model
SOURCE → OBSERVATION → ENTITY → EVENT
  • Single Postgres schema per tenant
  • Append-only, bitemporal observations
  • No ETL silos, no bespoke pipelines
  • Versioned migrations, audit-by-design
/03
Agent triad
ADJUDICATOR · INJECT · AFTER-ACTION
  • Anthropic Claude with structured outputs
  • Every claim cites a source observation
  • Suggestion mode only — nothing publishes
  • Faculty-grade override on every output
/04
Artifacts
SESSIONS · CASES · REPORTS
  • Multiplayer sessions with role asymmetry
  • Cases bind events across time + space
  • Signed PDF reports, base-rate cited
  • Five-day delivery window from session close
DEPLOYMENTSingle-tenant Postgres + Cloud Run. One database per engagement; zero cross-tenant query paths.
AUTHOIDC + per-role row-level security. Faculty bypasses asymmetry views; everyone else is enforced by the database.
EXTENSIBILITYNew data source = one ingestor + one schema migration. The canonical model is stable; the edges grow.
07b / UNDER THE HOOD

Boring choices, in service of the work.

The stack is opinionated and unfashionable on purpose. Every piece earns its place by being the simplest thing that holds up under audit and at scale. No frameworks adopted because of GitHub stars.

FRONTEND
Next.js 15React 19Tailwind v4MapLibre GL
BACKEND
Next.js API routesPayload CMSExpress (Node)TypeScript
DATABASE
Postgres 17PostGISpgvectorpg_trgm
AI
Anthropic Claudetiered routingstructured tool usegrounded citations
DEPLOY
Cloud RunCloud SQLeurope-west6 (CH)GitHub Actions CI
AUDIT
Append-only observationsEd25519 provenancePer-claim citationsRow-level security
Read the full architecture →
07 / IN ROOM

What serious use looks like.

insigz is used by people under time pressure with consequence. The room is concentrated, calm, expert — not theatrical. The interface respects the operator's attention; the tooling earns trust by being precise.

A wide cinematic photograph of three teams working in separate areas of a modern academic building

Three cells. One world. Each sees a different slice.

BALTIC · 17.02.2027
PHASE 2 / 3 · EXECUTE
47:23 REMAINING
12 PLAYERS · 3 CELLS
Working session in progress — multiple groups, screens, mid-discussion
Working session · mid-phase
Close portrait of a faculty member reviewing the approval queue on a tablet
Faculty review · approval queue
A printed after-action report lying open on a wooden desk in golden hour light
After-action report · signed PDF
07c / WHAT'S DIFFERENT

Three things we are not.

Said up front, so the next conversation starts at the right altitude. None of these are insults to anyone building in the adjacent spaces; they are simply not what insigz is.

/01

Not a defense product.

No classified data. No SCIFs. No sovereign sales. Built for commercial, journalistic, academic, and policy customers. That's a deliberate line, not a limit we'll quietly cross later.

/02

Not a SaaS dashboard.

No self-serve signup. No per-seat pricing. No shared multi-tenant pool. Each customer's deploy is single-tenant from day one — their own Cloud Run service, their own Postgres, their own Claude budget.

/03

Not a chatbot with maps.

Every Claude response is grounded in retrieved observations via tool use. Citations are clickable. Wrong answers are traceable. The LLM never hallucinates a fact past the model.

08 / ENGAGE

Three ways to put insigz to work.

Bespoke installations, intelligence-as-a-service, and structured session deliveries. No SaaS, no procurement marathons, no learn-our-platform tax. We deliver a working capability against a fixed scope.

/A · BESPOKE INSTALL

One tenant. One problem.

A scoped deployment of insigz for a single team, region, or use case. We bring the data fusion, the agents, the model. Hosted by us or by you.

On request
FIXED SCOPE
/B · INTELLIGENCE AS SERVICE

An analyst, augmented.

Recurring intelligence deliverables — sanctions enforcement briefs, shadow-fleet trackers, infrastructure risk monitors — drafted by the agents, signed by a named analyst on our side.

On request
12 MONTHS · MULTI-YEAR OPTIONAL
/C · SESSION DELIVERY

One day. One scenario.

A turnkey multiplayer session for an academy, an executive team, a policy workshop. Canonical scenarios available; bespoke scenarios on request.

On request
ONE DAY

Make the world
queryable.

CONTACT
hello@insigz.com
BASED IN
Luzern · Switzerland
RESPONSE
Within two business days.