Skip to content
EdgeQuake

EdgeQuake Feature Registry

Product: v0.23.0 · Contract: openapi.snapshot.json · Spec ops: Ingestion cancel & fairness

EdgeQuake Feature Registry

Section titled “EdgeQuake Feature Registry”

This file maintains traceability between code features, business requirements, and shipped releases (v0.11–v0.23).

Index

Section titled “Index”
Feature ID Description Status Spec / Release
FEAT-0001 Tenant Workspace Quota Management Completed SPEC-0001 / #133
FEAT-0002 Knowledge Injection (Glossaries & Synonyms) Completed SPEC-0002 / v0.8.0
FEAT-0003 Explainability & Model Picker (config chain) Completed SPEC-043 / v0.17.0
FEAT-0004 Graph Edge Labels Planned SPEC-0004 / #91
FEAT-0005 Custom Entity Configuration Completed SPEC-0005 / v0.8.0
FEAT-006 Unified Streaming Response Protocol Completed SPEC-006 / #56
FEAT-007 Vector Storage SQL Pre-Filtering Completed SPEC-007
FEAT-008 Explicit Provider/Model Transparency in UI Completed MISSION-01 / v0.9.19
FEAT-009 Document Deletion Correctness Completed MISSION-02 / v0.9.19
FEAT-010 Configurable PDF Parser Backend (Vision/EdgeParse) Completed MISSION-03 / v0.10.0
FEAT-011 Vision PDF Ingest & Side-by-Side Viewer Completed SPEC-047 / v0.17.0
FEAT-012 Real-Time Pipeline Progress (WS bridge) Completed SPEC-048 / v0.17.0
FEAT-013 Deletion & Reprocess Progress Parity Completed SPEC-050 / v0.17.0
FEAT-014 GraphRAG / Hybrid RAG Ops & ACC Science Completed SPEC-046 / v0.16.0
FEAT-015 OpenAPI-Native API Explorer Completed SPEC-035 / v0.15.x
FEAT-016 Mistral First-Class Provider Completed v0.11.0
FEAT-017 Embedding Progress Reporting Completed #197 / v0.11.3
FEAT-018 Runtime Auth Secure by Default Completed SPEC-027 / v0.13.x
FEAT-019 Documents List & Mix-Scale Perf Gates Completed SPEC-054 / v0.18.0
FEAT-020 Claim/Lease Delivery & Convert→Ingest SSOT Completed SPEC-057 / v0.19.0
FEAT-035 OpenAPI Explorer (WebUI implementation) Completed SPEC-035 — code marker
FEAT-094 Standalone PDF→Markdown Parse API Completed SPEC-094 / v0.23.0
FEAT-096 Multi-Language Extraction (workspace-scoped) Completed SPEC-096 / v0.23.0
FEAT-101 Wizard Onboarding & Setup API Completed SPEC-101 / v0.23.0
FEAT-102 Custom Entity Type Colors (workspace graph) Completed SPEC-102 / v0.23.0
FEAT-103 LightRAG-Parity LLM Cache Completed SPEC-103 / v0.23.0
FEAT-126 Provider KV / Prompt Cache Completed SPEC-126

Feature Definitions

Section titled “Feature Definitions”

FEAT-0002 — Knowledge Injection

Section titled “FEAT-0002 — Knowledge Injection”

Issue: #131
Spec: specifications/0002_knowledge_injection_issue_131
Released: v0.8.0 (2026-04-03)
Status: ✅ Completed

Problem: Domain-specific acronyms (OEE, NLP) and synonyms are unknown to the embedding model. Queries for “OEE” miss documents that say “Overall Equipment Effectiveness”, degrading retrieval quality.

Solution: Workspace owners inject glossary definitions as named entries. These are processed through the standard entity-extraction pipeline, enriching the knowledge graph. At query time, injection entities expand the query terms. Injection entries are never shown as source citations.

API Surface:

  • PUT /api/v1/workspaces/:id/injection — create/replace text injection
  • PUT /api/v1/workspaces/:id/injection/file — upload file injection
  • GET /api/v1/workspaces/:id/injections — list all entries
  • GET /api/v1/workspaces/:id/injections/:injection_id — get detail (plural path)
  • PATCH /api/v1/workspaces/:id/injections/:injection_id — update name/content
  • DELETE /api/v1/workspaces/:id/injections/:injection_id — delete + cascade cleanup

UI: /knowledge page with list, add dialog (text/file tabs), detail page, inline edit, delete confirmation.

FEAT-0003 — Explainability & Model Picker (SPEC-043)

Section titled “FEAT-0003 — Explainability & Model Picker (SPEC-043)”

Spec: specs/043-update-edgequake-llm
Released: v0.17.0 (2026-07-14)
Status: ✅ Completed (was Planned under legacy SPEC-0003)

Capabilities:

  • Unified ModelPickerPanel across workspace, query, and settings
  • Server-side model search: GET /api/v1/models/search
  • Provider Status Hub with auth_kind and remediation hints
  • Config explainability panel — effective provider/model resolution chain
  • Application attribution API for downstream LLM request labeling
  • Bundled models.toml with runtime override paths

API Surface: /api/v1/models/search, /api/v1/settings/*, /api/v1/config/effective

FEAT-0005 — Custom Entity Configuration

Section titled “FEAT-0005 — Custom Entity Configuration”

Issue: #85
Spec: specifications/0005_custom_entity_config_issue_85
Released: v0.8.0 (2026-04-03)
Status: ✅ Completed

Workspace-scoped entity_types with preset-driven and custom configuration, normalized and injected into extraction prompts per workspace.

FEAT-010 — Configurable PDF Parser Backend

Section titled “FEAT-010 — Configurable PDF Parser Backend”

Released: v0.10.0 (2026-04-10)
Status: ✅ Completed

Runtime PDF extraction backends: vision (VLM), edgeparse (CPU), edgeparse-ocr (EdgeParse + Tesseract), and auto. Resolution: per-upload → workspace default → EDGEQUAKE_PDF_PARSER_BACKEND env → vision.

FEAT-011 — Vision PDF Ingest (SPEC-047)

Section titled “FEAT-011 — Vision PDF Ingest (SPEC-047)”

Spec: specs/047-rag-evaluation
Released: v0.17.0
Status: ✅ Completed

  • PDF → Markdown via vision LLM (page-level rendering)
  • Per-page progress via WebSocket /ws/progress/{track_id}
  • Side-by-side PDF + Markdown viewer
  • Visual asset extraction (document_mm_assets, migrations 084/085)

FEAT-012 — Real-Time Pipeline Progress (SPEC-048)

Section titled “FEAT-012 — Real-Time Pipeline Progress (SPEC-048)”

Spec: specs/048-improve-ux
Released: v0.17.0
Status: ✅ Completed

  • spawn_pipeline_ws_bridge forwards pipeline events to WS/SSE clients
  • Pipeline status dialog with per-stage timing
  • track_id correlation upload → pipeline → completion
  • Progress endpoint: /ws/progress/{track_id} (not legacy /rag/*)

FEAT-013 — Deletion & Reprocess Progress (SPEC-050)

Section titled “FEAT-013 — Deletion & Reprocess Progress (SPEC-050)”

Spec: specs/050-pipeline-and-delete
Released: v0.17.0
Status: ✅ Completed

  • Delete document shows stage-by-stage progress (graph / vector / KV cleanup)
  • Reprocess parity with structured progress feedback
  • All pipeline stages surfaced with human-readable labels

FEAT-014 — GraphRAG / Hybrid RAG Ops (SPEC-046)

Section titled “FEAT-014 — GraphRAG / Hybrid RAG Ops (SPEC-046)”

Spec: specs/046-graphrag-study
Released: v0.16.0
Status: ✅ Completed

Fail-closed HNSW readiness, PPR-default graph walks, bipartite dual-node retrieval, ACC CI gate, ops runbooks.

FEAT-018 — Runtime Auth Secure by Default

Section titled “FEAT-018 — Runtime Auth Secure by Default”

Released: v0.13.x (SPEC-027 hardening)
Status: ✅ Completed

  • auth_enabled: true by default when unset
  • EDGEQUAKE_DEV_MODE=true opt-out for local make dev
  • Fail-closed middleware on versioned API when auth enabled
  • WebSocket auth rejects missing token when auth enabled

FEAT-019 — Storage/Query Performance Gates (SPEC-054)

Section titled “FEAT-019 — Storage/Query Performance Gates (SPEC-054)”

Spec: specs/054-fix-bugs-17
Released: v0.18.0
Status: ✅ Completed

Documents list perf gate, Mix-scale query budgets, batch lineage SQL, stable track_id across upload → WS progress.

FEAT-020 — Claim/Lease Delivery & Convert→Ingest (SPEC-057)

Section titled “FEAT-020 — Claim/Lease Delivery & Convert→Ingest (SPEC-057)”

Spec: specs/057-pipeline-reliability
Released: v0.19.0 (2026-07-17)
Status: ✅ Completed

Delivery SSOT:

  • Workers claim via FOR UPDATE SKIP LOCKED + leases; NOTIFY is wake-only
  • IngestionStatusMapper → API display_status / ui_phase on DocumentSummary
  • PDF Cancelled status (never maps cancel → Failed)
  • Convert (TaskType::PdfProcessing) then ingest (TaskType::Insert) with markdown checkpoint
  • Cancel facade: POST /api/v1/tasks/{track_id}/cancel
  • Multi-replica: EDGEQUAKE_REPLICAS + queue-metrics observability

Ops: Ingestion cancel & fairness

FEAT-035 — OpenAPI-Native API Explorer

Section titled “FEAT-035 — OpenAPI-Native API Explorer”

Spec: specs/035-api-explorer
Status: ✅ Completed

WebUI /api-explorer driven by OpenAPI snapshot with auth token and workspace base URL injection (@implements FEAT-035 in code).

FEAT-094 — Standalone PDF→Markdown Parse API (SPEC-094)

Section titled “FEAT-094 — Standalone PDF→Markdown Parse API (SPEC-094)”

Spec: specs/94-api-markdown
Released: v0.23.0 (2026-08-02)
Status: ✅ Completed

Stateless POST /api/v1/parse (multipart or raw application/pdf) converts PDF→Markdown without document residue:

  • GET /api/v1/parse/backends — list available parse backends
  • GET /api/v1/parse/jobs/{id} — poll async parse jobs (in-memory TTL)
  • Sync ceiling 15 pages + 20 MiB; async up to 1000 pages (Prefer: respond-async)
  • Returns Markdown + timing/cost metrics only

FEAT-101 — Wizard Onboarding & Setup API (SPEC-101)

Section titled “FEAT-101 — Wizard Onboarding & Setup API (SPEC-101)”

Spec: specs/101-wizard-mode-tenant-workspace
Released: v0.23.0 (2026-08-02)
Status: ✅ Completed

First-run wizard onboarding: setup API + context selector + provider/workspace bootstrap.

FEAT-103 — LightRAG-Parity LLM Cache (SPEC-103)

Section titled “FEAT-103 — LightRAG-Parity LLM Cache (SPEC-103)”

Spec: specs/103-llm-cache
Released: v0.23.0 (2026-08-02)
Status: ✅ Completed

Unified keyword + answer LlmResponseCache (L1 memory + L2 public.llm_cache):

  • Master switch EDGEQUAKE_LLM_CACHE defaults on; overrides EDGEQUAKE_KEYWORD_CACHE / EDGEQUAKE_QUERY_ANSWER_CACHE
  • Acc pins cache off (EDGEQUAKE_LLM_CACHE=0) for fair cold peers
  • Proof: make spec103-llm-cache-proof

FEAT-126 — Provider KV / Prompt Cache (SPEC-126)

Section titled “FEAT-126 — Provider KV / Prompt Cache (SPEC-126)”

Spec: specs/126-provider-kv-cache
Status: ✅ Implemented

Per-provider prompt/KV cache policy (default on). Distinct from SPEC-103 response cache:

  • Native OpenAI (OpenAIProvider, including official-OpenAI proxies) / Azure: prompt_cache_key plus GPT-5.6 explicit breakpoints (prompt_cache_options / prompt_cache_breakpoint). Support is learned from structured error.param 400s — not from model-name or hostname parsing. OpenAIProvider::compatible never sends GPT-5.6 fields.
  • Mistral / NVIDIA / OpenAI-compatible: prompt_cache_key only (eq:{role}:{provider}:{model})
  • Anthropic: cache_control on stable system prefix; TTL EDGEQUAKE_PROMPT_CACHE_TTL (5m/1h)
  • OpenRouter: cache_control + prompt_cache_key + session_id (sticky routing, Aug 2026)
  • Bedrock Converse: cachePoint after system blocks (protocol, not model-id sniffing)
  • Gemini / Ollama: layout-only (implicit prefix reuse; Gemini cachedContents when the flag is on)
  • Mix/answer + extract/glean/keyword/summarize/judge: stable instructions first, dynamic content last
  • Acc keeps this on (does not skip generation / does not change answers)
  • Highest-volume win is extract/glean (many chunks × same system). Mix KV is small unless Mix instructions exceed vendor min tokens.
  • E2E (CI): e2e_spec126_prompt_cache — Native explicit fields, Compatible omits them, structured param 400 retry. Live OpenAI cached_tokens proof is #[ignore] (EDGEQUAKE_LIVE_PROMPT_CACHE=1). Vendor APIs verified August 2026.

FEAT-096 — Multi-Language Extraction (SPEC-096 / GH-352)

Section titled “FEAT-096 — Multi-Language Extraction (SPEC-096 / GH-352)”

Spec: specs/096-multi-language-extraction
Issue: #352
Status: ✅ Implemented (v0.23.0)

Workspace-scoped extraction_language for KG entity/relationship natural-language output:

  • Allowlist aligned with LightRAG SUMMARY_LANGUAGE / SUPPORTED_LANGUAGES
  • Resolve: workspace metadata → EDGEQUAKE_EXTRACTION_LANGUAGE → English
  • Production JSON extractors inject a shared language instruction (JSON keys stay English)
  • WebUI WorkspaceExtractionLanguageCard beside Entity Types; future ingestions / reprocess only
  • Language-aware entity-type presets (LAW-L6): known presets remap to localized UPPERCASE tokens (e.g. French PERSONNE); custom lists stay as-is
  • OpenAPI + Playwright e2e screenshots under specs/096-multi-language-extraction/e2e/screenshots/

Ops note: Multilingual corpora often need a multilingual embedding model for retrieval quality (TrustGraph guidance).

FEAT-102 — Custom Entity Type Colors (SPEC-102)

Section titled “FEAT-102 — Custom Entity Type Colors (SPEC-102)”

Spec: specs/102-custom-entity-type-colors
Status: ✅ Completed (v0.23.0)

Workspace-scoped entity_type_colors for knowledge-graph visualization:

  • Persist { "PERSON": "#3b82f6", ... } in workspace metadata via create/update APIs
  • Single WebUI resolver (resolveEntityTypeColor) over expanded defaults
  • EntityTypeSelector + graph legend color pickers; entity-type mode only
  • Hex #RGB / #RRGGBB validation; max 50 entries
  • Gates: entity-type-colors.test.ts, spec102_entity_type_colors_persist, e2e/spec102-entity-type-colors.spec.ts

Release Map (v0.11 → v0.23)

Section titled “Release Map (v0.11 → v0.23)”
Version Date Highlights
0.11.0 2026-04-27 Mistral first-class provider
0.11.3 2026-05-06 Embedding progress; B2B header propagation; pipeline timeout env vars
0.12.0 2026-05-06 Vision image attachments; auth token expiry UX
0.13.x 2026-07 Auth secure by default; OIDC paths
0.14–0.15 2026-07 OpenAPI explorer; migration tooling
0.16.0 2026-07-10 SPEC-046 GraphRAG ops + ACC science
0.17.0 2026-07-14 SPEC-043 model picker; SPEC-047 vision; SPEC-048/050 progress
0.18.0 2026-07-16 SPEC-054 perf gates; OpenAPI snapshot freshness
0.19.0 2026-07-17 SPEC-057 claim/lease; convert→ingest; cancel fairness
0.20.0 2026-07-21 LightRAG Mix arms; Drawing display names; vision ingestion reliability
0.21.0 2026-07-23 Query-API parity (074–085); D-30 multigraph arbiter; SPEC-083 closure
0.22.0 2026-07-26 SPEC-090 multi-pool + migrate CLI; M104/M105 cutovers; boot migrate split
0.23.0 2026-08-02 SPEC-091 relational data-layer cutover; SPEC-103 LLM cache; SPEC-094 parse API; wizard/onboarding

Last Updated: 2026-08-03
Total Features (indexed): 25
OpenAPI SSOT: edgequake_webui/openapi/openapi.snapshot.json