1.0.0.big.update

[Dvurechensky]

Hi!

Preserved From ai-reverse-engineering-main

• Flask web server with upload, job, status, chat, and chat-history routes.
• Headless Ghidra REST service on http://localhost:9090.
• OpenAI-compatible chat completions workflow.
• Ghidra tool calls for function listing, decompilation, xrefs, imports,
  strings, and artifact search.
• Browser chat with Markdown, Mermaid diagrams, and syntax-highlighted code.
• Demo assets and the basic "upload binary, wait for analysis, ask questions"
  workflow.

Core Additions

• Local-first LLM mode through Ollama.
• Preserved hosted/global mode through LLM_PROVIDER=openai.
• Custom OpenAI-compatible gateway mode through LLM_PROVIDER=custom.
• Shared model configuration in webui/llm_config.py for chat, type recovery,
  and rename recovery.
• Web Settings panel for switching AI provider, model, endpoint, translator,
  and custom translator field mapping at runtime.
• Runtime sidebar showing provider, model, LLM endpoint, Ghidra endpoint, and
  translator status.
• /config route for frontend runtime configuration.
• Safer job-id and filename validation across chat, status, recovery, and file
  routes.
• Local job deletion with cleanup of data/, chat history, recovery indexes,
  and generated recovered files.
• Deleted-job tracking so locally removed jobs do not immediately reappear from
  the Ghidra service.

Recovery Pipeline

• Added webui/recovery_engine.py as the deterministic recovery orchestrator.
• Added dedicated recovery modules for functions, structures, classes, model
  output, and renaming.
• Generates recovery_manifest.json with file descriptions, validity labels,
  counters, and machine-readable metadata.
• Generates recovered_symbols.h, recovered_stubs.cpp,
  recovered_functions.cpp, and optional .renamed.* variants.
• Extracts dynamic module hints, GetProcAddress-style pointers, helper names,
  MSVC decorated symbol owners, pointer/offset structure candidates, and class
  layout candidates.
• Wraps selected Ghidra pseudocode in draft C/C++ files for inspection.
• Adds optional AI type/class recovery through AI Types.
• Adds optional AI rename-map generation through AI Rename.
• Rename output is conservative, validated, and non-destructive: renamed files
  are written as separate variants.
• Added fallback rename inference when the model returns too few safe names.

Analysis UX

• Added a dedicated Analysis tab next to Chat.
• Added syntax-highlighted C/C++/JSON/Markdown editor for recovered files.
• Added recovery progress cards for Ghidra, artifacts, recovery, and AI passes.
• Added a Symbol Map with old name, recovered name, address, signature, filters,
  and click-to-jump navigation.
• Symbol Map now supports All, Draft, Renamed, Raw, and Missing
  filters.
• Symbol clicks open the best generated source file and search by address,
  original name, recovered name, and metadata comments.
• Known function names and large hex addresses inside the source editor are now
  clickable without breaking syntax highlighting.
• Added a per-function inspector backed by local Ghidra artifacts. It shows the
  resolved address, signature, source file, callers, inferred callees, related
  strings, and draft/rename status.
• Inspector actions can hand off a focused prompt to the AI chat for function
  explanation, rename review, or VC++ 2003 reconstruction.
• Focused function lines are highlighted in the editor gutter.
• Added collapsible analysis navigation so the code editor can use more width.
• Added editor Wrap mode for long decompiler lines.
• Added editor Focus mode with Esc exit for large files.
• Replaced raw character counts with compact line and file-size badges.
• Added custom scrollbars, loading skeletons, upload drop-zone polish, and a
  branded empty-state screen.
• Split the old monolithic frontend into webui/static/css/app.css and
  webui/static/js/app.js.

Translation

• Added webui/translator_config.py.
• Added optional LibreTranslate support.
• Added optional custom JSON network translator support for proxying DeepL,
  Google, Argos, or another service.
• Added /translate route.
• Added per-message Translate buttons in chat.
• Russian text targets English; other text targets Russian.
• Translation preserves code blocks, inline code, symbol names, and hex
  addresses as much as possible before sending text to the translator.

Branding And Documentation

• Added media/icon.ico.
• Added favicon route and visible sidebar brand icon.
• Rebuilt README.md and README.ru.md with Windows and Linux/macOS startup
  instructions.
• Added translator setup to both README files.
• Added FEATURES.md, FEATURES.ru.md, TODO.md, and TODO.ru.md.
• Added webui/static/README.md to document the frontend module layout.
• Added a fuller .env.example.
• Filled .gitignore for virtualenvs, logs, generated Ghidra data, chats, and
  recovered artifacts.
• Added Dockerfile, .dockerignore, and docker-compose.yml for a local stack
  with Flask, Ghidra REST, and dvurechensky/libretranslate-offline-ru-en-zh.
• Added configurable compose host ports: WEBUI_PORT, GHIDRA_PORT, and
  LIBRETRANSLATE_PORT.
• Added persistent webui/settings/runtime_settings.json for web-edited runtime
  settings.
• Added logs/ with .gitkeep, ignored runtime log files, and rotating
  Flask/Werkzeug file logging.

Fixes

• Fixed mojibake-prone Russian docs by rewriting UTF-8 .ru.md files.
• Fixed Werkzeug header encoding issue by keeping user-facing non-ASCII errors
  out of response headers.
• Improved local job loading when Ghidra is offline.
• Improved delete UX when local files are removed but upstream Ghidra delete is
  unavailable.
• Fixed chat overflow and large-message formatting.
• Fixed Symbol Map navigation for renamed functions and address-based jumps.
• Added static cache-busting for CSS and JS during development.
• Verified the compose stack: web UI, Ghidra /jobs, LibreTranslate
  /languages, and Flask /translate responded successfully.

Known Limits

• Ghidra REST is still an external service and must be started separately.
• Recovery output is evidence and draft material, not guaranteed compile-ready
  source.
• The UI does not yet edit provider settings at runtime; configuration is via
  environment variables.
• APK analysis is a roadmap item, not a working pipeline.
• Production deployment, auth, queues, quotas, and multi-user isolation are not
  included in this local workbench release.