Skip to content / 本文へ

PRODUCT / CONTEXT STUDIO

Explore, verify, and curate your organizational context graph

Connect extracted assertions, passages, business terms, and structured metadata on a bitemporal graph with human-in-the-loop review.

Bundata Context Studio is the unified workspace where document extractions, source passages, verified business definitions, and database metadata meet. Investigate relationships on an interactive graph canvas, resolve contradictions in prioritized review queues, and publish curated context sets for search and AI agents.

Context Studio · Bitemporal Investigation
True at: Now · As known at: Now
ExploreEvidenceQuality ReviewLibraryDataEnrich
Graph Canvas · 4 nodes · 3 edgesCollection: Vendor Contracts

MSA_Agreement_2026.pdf

Document · Version 3 · 40 pages

+ Expand

Contains passage → Smart Bite

§12.2 Termination · Page 4

Smart Bite

Extracted assertion → Verified

Notice period: 30 days

Evidence InspectorVerified anchor
Field:termination_notice_days
Value:30 days
Source span:p.4 · §12.2 · v3

Verified source quote:

p.4 §12.2 — “...either party may terminate upon thirty (30) days' prior written notice...”

See how teams investigate source lineage and verify extracted assertions

Walk through an active extraction contract, inspect field coverage, expand document nodes on the bitemporal graph, and verify extracted assertions against their exact source quote.

SIX INTEGRATED WORKSPACES

How Context Studio works in the Bundata platform

Context Studio (/platform/context-graph) brings together six specialized tabs over a shared bitemporal graph and permission boundary. Every view enforces live collection access checks and preserves exact source provenance.

Tab 1 · Explore

Interactive Graph Canvas & Discovery Chat

A bounded, tool-using chat agent sits beside a live context-graph canvas. Ask discovery questions across 1–5 collections using fused semantic and Hebrew/English BM25 retrieval, expand connections between entities and documents, and use @document, @tables, @rules, @history, and @sql mentions.

Tab 2 · Evidence

Lineage Investigation & Time-Travel Inspector

Trace any document, Smart Bite passage, or extracted assertion across bitemporal cursors ('True at' valid time and 'As known at' system time). Inspect exact source spans, document versions, page ranges, and verified source quotes in the Evidence Inspector.

Tab 3 · Quality Review

Prioritized Graph Quality Review Console

Resolve signals that require human judgment: uncertain entity merges or splits, stale superseded documents, claim contradictions, low-confidence facts, table reconciliation mismatches, and OCR issues. Every decision writes an audited, reversible bitemporal assertion.

Tab 4 · Library

Governed Business Context Library

Maintain immutable, versioned business terms (with Hebrew and English aliases), verified answers with inline [1] passage citations, instructions, and read-only SQL query templates. Citations are continuously checked against source document currency.

Tab 5 · Data

Structured Database & dbt Metadata

Connect structured database schemas, tables, views, columns, join paths, and dbt models into the same graph. Link business terms to columns and run approved, parameter-bound SQL templates directly from the studio.

Tab 6 · Enrich

Learning from Questions & One-Click Proposals

Detect knowledge gaps from real user questions and run bounded enrichment jobs that draft new terms, term updates, conflict alerts, and verified answers from changed documents—waiting in a review inbox for steward approval.

Built on verifiable evidence, not unexamined black boxes

Bitemporal Graph Engine

Query what was true in the business at a specific moment ('True at') alongside what the system knew at a given time ('As known at'). Reviewer corrections archive rejected assertions without erasing history.

Live Citation Currency

Library entries and assertions store exact UTF-16 spans and SHA-256 fingerprints against document versions. When a source document updates or expires, affected context is automatically flagged for re-anchoring.

Release Gates & Evaluation Promotion

Block workflow publication when critical contradictions or stale-document alerts exceed your age threshold, and promote reviewed corrections directly into Eval Studio datasets.

END-TO-END FLOW

From extracted fields to curated agent memory

Context Studio turns isolated document extractions into governed organizational memory that powers Search & Answer and Agent Studio.

  • Ingest documents and run schema-aware extraction to produce Smart Bites and typed assertions
  • Explore multi-document connections and entity relationships on the interactive graph canvas
  • Inspect bitemporal lineage and verify extracted values against highlighted source quotes
  • Adjudicate contradictions, entity merges, and table checks in the Quality Review queues
  • Approve cited business terms, instructions, and verified answers into the Context Library
  • Publish curated focus nodes into versioned Context Sets for downstream agents and workflows

Connected across the Bundata platform

Extraction Studio

Feed schema-aligned fields, Smart Bites, and table facts directly into the context graph.

Vector Catalog

Inspect collection health, active extraction contracts, and document-level lineage.

Governance & Lineage

Enforce access boundaries, audit logs, confidence monitoring, and release gates.

Frequently asked questions about Context Studio

Context Studio (/platform/context-graph) is Bundata's workspace for exploring, reviewing, and governing the knowledge graph and organizational memory built from your documents and data sources. It combines an AI-assisted graph exploration canvas, an evidence and lineage inspector, a quality review console, a business context library, structured metadata browsing, and automated context enrichment.

Context Studio uses a bitemporal graph. 'True at' filters facts and documents by when they were valid in the real world, while 'As known at' shows the graph and library state as it was recorded in the system at that timestamp. Changing cursors re-reads the graph deterministically without calling an AI model.

When entity resolution, claim verification, table reconciliation, or extraction confidence checks detect an issue, it enters a prioritized queue. A human reviewer inspects the source passages and PDF previews, then accepts, rejects, or edits the item with a mandatory reason. The decision is saved as an audited, reversible bitemporal assertion that re-extraction will never overwrite.

The Library holds steward-approved business terms (with English and Hebrew aliases), verified answers to recurring questions, agent instructions, and vetted read-only SQL templates. Every entry requires at least one verified passage citation and is checked automatically whenever source documents change.

No. Context Studio relies on stored documents, bitemporal graph assertions, versioned library entries, and saved exploration state. Dedicated non-retaining workspaces store no graph or library data and explicitly block Context Studio API routes.

Ready to explore your document intelligence graph?

Inspect source evidence, curate verified business context, and ground your AI systems with Context Studio.