Decisions you can defend,
at the speed of the business.
Generative BI powered by a live ontology of Kapiva. A Lead Agent plus specialist agents reason over your metrics, entities and conventions — and every recommendation is cited, every decision logged. No more black-box dashboards.
Built for Kapiva — for forecasting, growth and operational decisions.
One question. Four moves. A defensible answer.
Every chat is a planned trace through your ontology — never a hallucinated summary of a dashboard.
“Why did D2C revenue dip 18% last week? Will it recover by Friday?”
RCA pinpoints SKU-level checkout drop. Forecast projects partial recovery by Fri. Each claim linked to runs, queries, and source docs.
Dashboards answer 'what'. Tattva answers 'why'.
Generative BI on top of a live ontology of your business — so the engine doesn't search loose text, it navigates structure.
The Live Cognitive Ontology
Your business as a typed graph. Metrics, entities, conventions and decisions are modeled and linked — so when the Lead Agent asks “why is D2C down?”, it traverses connections, not strings.
- Metrics linked to their owning entities, time grains and conventions
- Entities — SKU, channel, region, customer cohort — typed and joinable
- Conventions — what “D2C revenue” means, who owns it, how it’s computed
- Decisions — past calls and outcomes, so the engine learns from history
- 30 dashboards, 0 answers. You still have to interpret.
- No memory of why a number means what it means.
- Decisions disappear into Slack threads and meeting notes.
- Ask in natural language. Get cited reasoning, not just charts.
- The ontology knows how Kapiva defines “D2C” — so the engine does too.
- Every decision lands in the ledger, linked back to its evidence.
Six agents. One coherent answer.
Specialists share tools, ontology, and a single audit trail — not chatbots stitched together.
Decomposes a metric anomaly into the drivers that moved it — by SKU, channel, geo.
Projects forward with stated assumptions. Shows confidence bands, not point bets.
Connects spend, exposure and conversion across channels — without dashboard lying.
Recommends levers (price, mix, allocation) with expected lift and tradeoffs.
Forward-looking what-ifs on cadence. Flags the leading signals before they tip.
Plans the path, picks specialists, composes the final answer with every step traced.
Built to be defended, not just believed.
Coverage you can see, citations on every claim, and the systems you already use — all wired into a single trail.
S/Se/A/D/I coverage
The ontology's coverage scorecard — Structured, Semi-structured, Auxiliary, Documents, Interactions — tells you exactly what the engine understands.
Citations on every claim
Each sentence in the answer links to the run, query, document chunk or decision that backs it. If you can’t trace it, the engine won’t say it.
- Inline tool-call rendering: see what was actually called
- Hypothesis ledger — accepted, rejected, deferred
- One-click jump from answer to evidence
Connected to everything we already have
BigQuery, Cloud SQL, Drive, call recordings, file uploads. The engine parses, embeds and links — so the ontology stays current.
- Structured: BigQuery datasets + Cloud SQL
- Unstructured: Drive, PDFs, call transcripts
- Bring-your-own tools via MCP-style adapters
Stop arguing about the dashboard.
Start defending the decision.
Sign in to open your workspace and ask the first question. Most teams reach a cited recommendation inside the first hour.