Skip to content

CHANGELOG.md

Latest commit

 

History

History
126 lines (88 loc) · 5.9 KB

CHANGELOG.md

File metadata and controls

126 lines (88 loc) · 5.9 KB

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

[2.1.0] - 2026-03-22

Changed

  • POS highlighting controls are now feature-gated via config.toml ([features].pos_highlighting) and default to hidden/off for minimal UI by default.

[2.0.0] - 2026-03-20

Added

  • Added optional text-only mode for both phases (media upload is no longer required).
  • Added multi-media upload support across Phase 1 and Phase 2 (single or multiple image/video files).
  • Added per-item media tagging with editable defaults (@imageN / @videoN) and tagged payload composition for multi-media requests.
  • Added a dedicated Manage media tags dialog with full-size preview and tag editing.
  • Added output editing dialogs for both Phase 1 and Phase 2, with edited content persisted to conversation context.
  • Added a large sidebar dialog editor for the System Prompt.
  • Added optional session request logging with JSON export download.
  • Added configurable hero notices via config.toml.
  • Added offline CI pytest coverage plus opt-in live smoke execution via --live and .env loading in tests.

Changed

  • Updated workflow terminology to Primary Analysis (Phase 1) and Refinement Loop (Phase 2).
  • Updated System Prompt behavior so the editable textbox is the request source of truth; preset dropdown now loads via explicit Load action.
  • Improved Request Transparency previews for multi-media payloads (media summaries and media-tag mapping).
  • Updated output copy actions to support both Copy Plain Text and Copy Markdown.
  • Updated phase headers and reasoning sections with Material icon labels.
  • Refined in-app notice content, including Xiaomi/OpenRouter update messaging.

Fixed

  • Fixed media-tag persistence to remain stable when media items are added/removed (signature-based mapping).
  • Improved setup reliability by bundling en_core_web_sm via dependency wheel and aligning guidance around uv sync.
  • Improved UI edit ergonomics by widening system prompt and phase-output edit dialogs.

Documentation

  • Updated README.md, AGENTS.md, docs/REFERENCE.md, and docs/TESTING.md to match current behavior and testing workflow.

[1.2.1] - 2026-03-13

Fixed

  • Fixed POS highlighting so enabling it no longer breaks markdown rendering in model outputs.
  • Updated highlighter behavior to be markdown-aware by preserving fenced code blocks, inline markdown spans, and line-prefix structure markers while highlighting plain text tokens.

[1.2.0] - 2026-03-12

Added

  • Added support for both image and MP4 video inputs in Phase 1 and Phase 2 analysis flows.
  • Added folder-based prompt preset loading with optional metadata sidecars.
  • Added a new initial prompt preset: video technical hybrid script.

Changed

  • Removed leftover debug reasoning-effort caption from the chat render flow.
  • Updated user-facing docs copy to generalize MP4 mentions to broader “video” terminology.

Documentation

  • Expanded AGENTS.md guidance for video workflow and prompt preset behavior.
  • Clarified optional spaCy POS-highlighting requirements in docs.

[1.1.1] - 2026-03-09

Changed

  • Made Phase 1 and Phase 2 prompt fields editable in the UI so users can tune instructions per run without leaving the workflow.
  • Cleaned up request transparency preview presentation for a clearer, more compact phase action experience.

[1.1.0] - 2026-03-09

Added

  • Provider presets via config.toml for endpoint/model selection (e.g., BytePlus/OpenAI/Gemini/OpenRouter)
  • Unified API key env fallback flow with LLM_API_KEY plus provider-specific key compatibility
  • Hero landing section (hero.md) with FrameLab branding and two-phase workflow guide
  • API Setup section wrapped in an expander for cleaner sidebar/UI layout
  • Per-phase Request Transparency preview panel showing provider, endpoint, model, reasoning setting, and compact payload preview
  • UI processing state controls to lock inputs/buttons during active analysis and prevent duplicate submissions
  • Hero badges (latest release, MIT license, Python 3.11+) and creator credit line

Changed

  • Request transparency previews are now rendered inline with each phase action panel for clearer workflow alignment
  • Setup/configuration docs now reflect provider preset workflow and unified env-key handling
  • Added requirements.txt for pip-based installs and set Poetry package mode to dependency-management-only (package-mode = false)

Fixed

  • Improved phase layout consistency by removing placeholder-based transparency rendering and placing action buttons after each inline preview

[1.0.0] - 2026-03-08

Added

  • Initial release of FrameLab, a lightweight multimodal AI web app for cinematic image analysis
  • Minimal dependencies: only streamlit and openai
  • OpenAI-compatible endpoint support via configurable Base URL
  • Hybrid configuration fallback: sidebar input → .env → config.py defaults
  • API key support with sidebar override priority
  • Externalized default system prompt via system_prompt.txt
  • Uploaded image previews for Phase 1 and Phase 2
  • Real-time streaming output in UI
  • Thought/reasoning stream shown in expandable section
  • One-click copy buttons for Phase 1 and Phase 2 results (plain text)
  • Token usage summary (input/output/total) when provided by endpoint
  • Responses API support with automatic Chat Completions fallback
  • Transport path display (Responses API or Chat Completions fallback)
  • Auto-disable Responses API when provider reports schema mismatch
  • Two-phase workflow:
    • Phase 1: Initial image analysis
    • Phase 2: Correction loop with new image + notes
  • Session-based conversation memory
  • Reasoning effort configuration (none/minimal/low/medium/high)
  • Error handling with underlying exception messages for debugging