Skip to content

Index Dashboard Models

  • Storage Breakdown

    DashboardIndexStorageBreakdown shows bytes across Postgres + Neo4j.

  • Costs

    DashboardIndexCosts estimates embedding, semantic KG, and figure-description costs.

  • Embedding Config

    DashboardEmbeddingConfigSummary summarizes vector storage settings.

Get started Configuration API

At-a-glance

Use dashboard stats to detect storage drift (e.g., pgvector index growth) and plan compaction windows.

Estimates

Some values (e.g., GIN/GIST index allocations) may be estimated when exact attribution is not possible.

Quotas

Track total storage vs. quotas per environment to avoid surprise outages.

Model Fields
DashboardIndexStorageBreakdown chunks_bytes, chunk_summaries_bytes, qdrant_points, qdrant_dense_vector_bytes, neo4j_store_bytes (nullable when unmeasured), neo4j_store_note, postgres_total_bytes, total_storage_bytes
DashboardIndexCosts total_tokens, embedding_cost, semantic_kg_cost, figure_description_cost, figures_described, total_cost, accounting (saved IndexRunAccounting for the live generation's run)
DashboardIndexStatusResponse lines, metadata, running, progress, current_file

Costs are saved observations, never repriced

Every cost field now comes from the accounting saved with the live generation's index run (resolved through the corpus manifest, not the newest summary on disk): the frozen pre-run estimate the run captured under its own config, plus the reconciled native spend when reconciliation has settled. Current configuration never reprices a past run — change the embedding model and the card still shows what the live generation actually cost. A corpus with no manifest-named run reports null for every component, and a missing or unpriced component keeps the total unknown rather than counting as zero. The accounting block carries the full census (admitted/in-flight requests, retained workers, coverage and pricing states, reconciliation reasons). See Native run accounting.

An unmeasured store is null with a reason, never a measured-looking zero

Neo4j 5 Community exposes no store-size source — dbms.queryJmx is gone in Neo4j 5, APOC core has no apoc.monitor.store, SHOW DATABASES carries no size column, and the data volume is a root-owned Docker volume the API process cannot read. When no measurement source exists, neo4j_store_bytes is null with neo4j_store_note saying why, the store contributes nothing to total_storage_bytes, and both Dashboard panels render the tile as n/a instead of a measured-looking 0 B (0.0% of total) beside a graph of thousands of nodes.

flowchart LR
    API["/index/stats"] --> UI["Dashboard Panels"]
import httpx
print(httpx.get("http://localhost:8000/index/stats?corpus_id=tribrid").json())

curl -sS "http://localhost:8000/index/stats?corpus_id=tribrid" | jq .

const stats = await (await fetch('/index/stats?corpus_id=tribrid')).json();
Metadata

DashboardIndexStatusMetadata.embedding_config aligns with the active embedding provider/model/dimensions.