Skip to content
EdgeQuake

Feature Tour

Released: v0.36.0 · Feature IDs and code anchors: Feature registry

Feature Tour

Section titled “Feature Tour”

This page lists what EdgeQuake can do today, in plain terms, with a link to learn more. It is for evaluators and new users.

%%{init: {"theme":"base","themeVariables":{"primaryColor":"#E0E7FF","primaryBorderColor":"#6366F1","primaryTextColor":"#1E1B4B","secondaryColor":"#D1FAE5","secondaryBorderColor":"#10B981","secondaryTextColor":"#064E3B","tertiaryColor":"#FEF3C7","tertiaryBorderColor":"#F59E0B","tertiaryTextColor":"#6B7A90","lineColor":"#7A889C","clusterBkg":"rgba(99,102,241,0.07)","clusterBorder":"#A5B4FC","noteBkgColor":"#FEF9C3","noteTextColor":"#422006","textColor":"#6B7A90","titleColor":"#6B7A90","signalColor":"#7A889C","signalTextColor":"#6B7A90","loopTextColor":"#6B7A90","edgeLabelBackground":"#F1F5F9","actorLineColor":"#94A3B8"}}}%%
%% eq-theme:v1
flowchart LR
    A["Upload: text, PDF, images"] --> B["Ingest pipeline"]
    B --> C["Knowledge graph and vectors"]
    C --> D["Six query modes"]
    D --> E["REST API, UI, MCP"]
%% eq-classes
classDef eqActor fill:#FCE7F3,stroke:#EC4899,color:#500724
class E eqActor

Read the chart from the left. Anything you upload goes through one pipeline into PostgreSQL. You query it through the API, the web UI or MCP.

Knowledge graph

Section titled “Knowledge graph”
  • Entity extraction. A model finds people, organizations, places, concepts and more. See Entity extraction.
  • Relationship mapping. Each link carries keywords and a description.
  • Gleaning. An optional second pass looks for missed entities. It is on by default for cloud providers and off by default for local ones. See Gleaning.
  • Community detection. Louvain clustering groups related entities. See Community detection.
  • Custom entity types. Choose a preset in the UI (General, Manufacturing, Healthcare, Legal, Research, Finance, or Blank), or define up to 20 types per workspace through the API.
  • Knowledge injection. Add glossaries, acronyms and synonyms to a workspace. See Knowledge injection.

Decision extraction (preview)

Section titled “Decision extraction (preview)”

A second extraction mode answers closed questions with a small local model (default tev1:0.8b on Ollama). Accepted answers enter the same graph. The chat-model extractor stays the default. Gate presets are uncalibrated, and this path has no SPEC-001 accuracy score. See Decision extraction.

Query engine: six modes

Section titled “Query engine: six modes”
Mode Best for
naive Simple fact lookups from chunk text
local Questions about one entity and its neighbors
global Broad, thematic questions
hybrid Local, global and naive arms interleaved
mix (default) Local, global and naive arms blended by weight or RRF
bypass Model only, no retrieval

This page gives no latency figures. Latency depends on your model, hardware and data size. See Hybrid retrieval and Product limits.

Retrieval details

Section titled “Retrieval details”
  • Graph walk. The default walk is breadth-first (bfs). Set EDGEQUAKE_GRAPH_WALK=ppr to use Personalized PageRank.
  • Bipartite entity and chunk pick. Local, global and mix use links between entities and chunks to choose chunks.
  • HNSW readiness check. With EDGEQUAKE_HNSW_PARTIAL_BY_WORKSPACE on, /ready fails while a vector table lacks its HNSW index. Warm the index with POST /api/v1/admin/ann/warmup.
  • Intent-gated arms. mix and hybrid can skip arms that do not help the question.
  • Failed-chunk retry. Extraction failures are stored, listed and can be retried.
  • Faithfulness sampling. A heuristic check, plus an optional model judge (EDGEQUAKE_FAITHFULNESS_JUDGE).
  • Accuracy gate. make spec046-acc writes a deterministic report without an API key.

Runbooks: SPEC-046 ops runbooks.

PDF pipeline

Section titled “PDF pipeline”

PDFs run in two tasks: convert to Markdown, then ingest.

Backend What it does Needs a model?
vision (default) A vision model reads each page as an image Yes
edgeparse Fast text extraction on the CPU No
edgeparse-ocr EdgeParse plus Tesseract for scanned tables No
auto Automatic routing between backends. Opt-in only; never the default Depends on the route

Tables and multi-column layouts are handled best by the vision backend. See PDF processing.

Production features

Section titled “Production features”
  • REST API. OpenAPI 3.1, streaming responses, batch upload, health and readiness probes.
  • Multi-tenant. Tenant and workspace isolation, enforced in PostgreSQL with row-level security.
  • Auth and audit. Built-in login, API keys, enterprise SSO (v0.30.0) and audit logging.
  • PostgreSQL 16, 17 and 18. With pgvector and Apache AGE.
  • Explicit schema. Schema changes run through edgequake migrate. By default the server does not start while the schema is pending (EDGEQUAKE_SCHEMA_GATE=fail).
  • Multi-arch images. linux/amd64 and linux/arm64, published to GHCR on each release.
  • MCP. Expose EdgeQuake to AI agents through the Model Context Protocol server.
  • Web UI. Next.js and React 19, with live progress and an interactive graph view.
  • Providers. Ollama, OpenAI, Anthropic, LM Studio, oMLX and other OpenAI-compatible servers. See Providers.