Workspace Management System (WMS) + Context Management System (CMS)

Case Workbench

The sovereign workspace console — case directory isolation, granular active-context control, and a live multi-dimensional entity graph that makes every case document instantly retrievable and explorable.

Document Ingestion Case Wiki
Isolated
Per-Case Directories
3 States
Context Control Levels
Hybrid
Lexical + Neural Retrieval
3D Graph
Case Entity Viewer
Left Panel Free in Perpetuity

Case Evidence & Matter Wiki Are 100% Free

Ingest trial documents, manage the Active-Context Matrix, and compound your matter wiki with zero subscriptions. Case files are stored in local SQLite databases that you own completely.

11.2M Diligence Hub
Case Manager

Sovereign Case Directory Management

Each CIRP case lives in a fully isolated local directory structure. No data bleeds between cases.

IsolationAir-Gapped

Isolated Case Workspaces

Each case gets its own isolated directory, encrypted local data store, vector index, and structured case configuration. Opening a new case activates its workspace context in the sidebar — no data bleeds between concurrent cases.

Per-case Store Zero cross-case leakage
Policy ControlConfig Hub

Workspace Policy Configuration

Each case carries a structured workspace policy that stores: ingestion modes (interactive / supervised), statutory milestone schedules, active context document selections, and session state snapshots used for context compaction.

Unified Policy Milestone scheduler
CIRP TimelineAuto-Scheduled

Milestone Self-Wake

On case initialization, the background daemon registers each statutory deadline (T0 → T330) as an automated wake event. When a statutory milestone passes, a structured notification card is dropped in the Case Action Inbox with the statutory reference and recommended action.

T0 → T330 tracked Inbox alert

Context Control

Active-Context Control Matrix

A granular control panel that maps every ingested document to one of three retrieval states, optimizing how information reaches the local model's memory budget.

Exclude
Excluded from all retrieval
The document's chunks are filtered out before lexical and vector retrieval. Useful for superseded filings, duplicate documents, or irrelevant financial periods that would pollute the context budget.
Insights Only
High-level summary retrieval only
Only the document's synthesized summary page is eligible for retrieval — not individual raw chunks. Preserves context budget while retaining the core facts of large background documents.
Full Context
Complete chunk-level retrieval
All chunks from the document are candidates in lexical and sovereign vector retrieval. Use for the most critical case documents — the Resolution Plan, Form H, and key NCLT orders that require granular citation.

Persistence & Token Budget

Selections are persisted across sessions in the workspace policy. The Concepts panel shows a live dynamic token count calculation so the practitioner can see exactly how many tokens the current selection will consume from the local model budget before sending any query.


Concepts & Entities

Hybrid Lexical & Vector Retrieval & Case Entity Graph

Hybrid retrieval pipeline combining sub-millisecond lexical keyword search with sovereign semantic reranking — paired with an interactive 3D entity graph for visual case exploration.

Hybrid RAGRRF Fusion

Dual-Stage Lexical & Vector Search

High-speed lexical search runs first to select candidate chunks via exact keyword match. Sovereign vector similarity is then computed on-the-fly over dense vector projections to rerank. Results are merged via Reciprocal Rank Fusion (RRF) for optimal precision.

<1ms lexical pass Dense rerank
Drift-FreeNatural Keys

Drift-Safe Chunk Matching

Candidates are matched across retrieval stages using composite natural keys (document identifier, section header, chunk sequence). This guarantees drift-free retrieval when case repositories are synchronized or backed up.

Natural key join Zero drift failures
Graph Viewer3D Interactive

3D Case Entity Graph

The Case Graph Viewer renders typed directional edges between case entities (creditors, applicants, filings, transactions). Edge types: related_party, contradicts, supersedes, cites. Contradiction edges surface discrepancies between filings visually.

Typed edges Contradiction flags

Entity Graph Edge Legend

contradicts — mismatched amounts across two filings
related_party — creditor → resolution applicant link
cites — document referencing another filing
supersedes — amended filing over original