Skip to content

release-record.md

Latest commit

 

History

History
125 lines (89 loc) · 4.83 KB

release-record.md

File metadata and controls

125 lines (89 loc) · 4.83 KB

Release Record

2026-03-23

Completed This Round

  • Refined frontend information architecture across Workspace / Insights / Collections / Archive / Library
  • Added dedicated Chat navigation entry to separate QA from document operations
  • Added in-page 专注对话 quick action on Documents page for one-click jump to full chat mode
  • Expanded dedicated chat panel area and feed height to improve readability on wide screens
  • Reduced overall UI typography scale for better visual density and consistency
  • Rebuilt frontend with npm run build after the structural and style changes

Why This Matters

  • QA flow is no longer constrained by the two-column document layout
  • Workspace now supports both operation mode (Documents) and focused conversation mode (Chat)
  • The revised typography and spacing reduce visual fatigue in long sessions

Current State

  • Core QA workflow is stable in both entry points:
    • Workspace -> Documents -> 专注对话
    • Workspace -> Chat
  • Frontend build passes after the UI structure update

Known Gaps

  • No in-page hide documents toggle yet; current strategy is dedicated Chat page
  • Full responsive manual pass for all redesigned pages is still pending

Recommended Next Step

  • Add one lightweight responsive verification round (desktop + tablet + mobile) for all workspace tabs
  • Then decide whether to add optional in-page Hide Docs toggle in Documents view

2026-03-20

Completed This Round

  • Passed real local Dify smoke for health -> auth -> knowledge base -> upload -> ingestion -> QA / SSE
  • Manually validated the frontend workbench on http://localhost:3001
  • Fixed local frontend CORS for localhost:3001 / 127.0.0.1:3001
  • Fixed Dify Completion API payload compatibility by sending query in both top-level payload and app inputs
  • Fixed dataset creation compatibility when knowledge base description is empty
  • Added an in-repo OCR adapter service under apps/ocr
  • Switched OCR runtime to RapidOCR + opencv-python-headless
  • Wired OCR into deploy/docker-compose.yml
  • Fixed backend OCR multipart compatibility with the OCR service contract
  • Passed real OCR smoke for PDF -> OCR -> Dify ingestion
  • Added Linux-executable smoke script smoke.sh
  • Added Linux deployment rehearsal wrapper rehearse-linux.sh
  • Verified streamed answers and source rendering in the browser
  • Verified new ingestion tasks complete successfully after the backend fixes
  • Added owner-only failed document delete and failed ingestion task retry in backend APIs
  • Added frontend recovery actions for failed document cleanup and failed task retry

Current State

Core product flow currently provides:

  • login / registration
  • knowledge base creation and selection
  • document upload
  • ingestion task tracking
  • Dify dataset / document synchronization
  • OCR-backed PDF ingestion
  • streamed QA with retrieved sources
  • Linux-executable smoke / rehearsal package

Still pending:

  • real Linux host deployment rehearsal
  • bulk cleanup / bulk retry UX for historical failures

Recommended Next Step

  • Execute bash scripts/rehearse-linux.sh on the target Linux host
  • Then decide whether bulk cleanup UX is needed beyond the current single-item recovery path
  • If deployment hosts are network-constrained, harden image build caching before release

2026-03-18

Completed This Round

  • Confirmed formal technical baseline: Java 21, Dify self-hosted, same-host split containers, MinIO
  • Delivered document upload and ingestion task tracking
  • Delivered Dify dataset/document API integration
  • Delivered OCR orchestration for PDF ingestion
  • Delivered QA / chat SSE with source-return and refusal behavior
  • Delivered frontend business workbench with upload, task, and QA panels
  • Added deployment smoke script for health, auth, KB, upload, ingestion, and QA validation
  • Aligned Docker runtime env forwarding with the delivered Dify/OCR feature set
  • Rehearsed the project Docker stack locally and confirmed actuator/health = UP
  • Captured the real external dependency blocker: Dify connect failure at http://host.docker.internal:8088/v1/datasets
  • Fixed authenticated SSE handling for async dispatch
  • Updated env templates, deployment boundary, and project docs
  • Passed frontend lint / build
  • Re-passed backend test / lint / build

Current State

Backend currently provides:

  • Auth
  • Knowledge base basic management
  • Document upload
  • Ingestion task records
  • Dify dataset/document synchronization
  • PDF OCR ingestion path
  • Knowledge-base-scoped streaming QA

Still pending:

  • Real external-service smoke tests

Recommended Next Step

  • Run real Dify / OCR smoke tests
  • Then do one Linux deployment rehearsal