paper-idea-memory

# Tool List A local knowledge incubator for **personal scientific research work** - gathering **Paper Research, Idea Flash, Experiment Notes, Concept Dictionary** together, can be connected to any large model (OpenAI / DeepSeek / Gemini / Claude / AG-UI / any OpenAI compatible API) for in-depth dialogue. > Inspired by SillyTavern's multi-model routing + Notion's knowledge organization + scientific research paper reading process. ## Features - **Multi-Entity Knowledge Base**: Paper · Idea · Note · Experiment · Reading-log · Concept - **Multi-Protocol LLM Routing**: OpenAI / DeepSeek / Anthropic Claude / Google Gemini / AG-UI / any OpenAI compatible endpoint - **Chat Tavern Style**: multi-session, message editing/regeneration, system prompt presets, multi-model side-by-side comparison - **Day / Night / Follow System** theme switching - **Local Encryption**: store API Key (AES + machine fingerprint) - **Pure File JSON Storage**, zero database dependency, reserved SQLite upgrade interface - **Idea Kanban**: sprout → research → verification → drafting → produced - **Paper Workflow**: arXiv / DOI / PDF import, BibTeX export, automatic summary ## Roadmap - [x] **Phase 1** (current): scaffolding + backend + multi-protocol Provider routing + basic conversation page - [ ] **Phase 2**: Paper import (arXiv / PDF), knowledge card CRUD, Idea Kanban - [ ] **Phase 3**: tavern-level conversation experience (multi-session, branching, presets, comparison mode, formula rendering) - [ ] **Phase 4**: BibTeX import/export, citation chain tracking, visual relationship graph ## Quick Start ```bash cd paper-idea-memory pip install -r requirements.txt bash start.sh # macOS / Linux # Windows: start.bat ``` Open browser to <http://127.0.0.1:3200>, click ⚙ **Connection Management**: 1. Select provider type (OpenAI / DeepSeek / Gemini / Claude / AG-UI / Custom-OpenAI) 2. Fill in API Key / base_url / enterprise WeChat user-token (AG-UI scenario) 3. Select default model 4. Save → start conversation ## Project Structure ``` paper-idea-memory/ ├── server.py # Flask entry (simplified, business split to routes/) ├── requirements.txt ├── start.sh / start.bat ├── config.example.yaml # default configuration template │ ├── lib/ │ ├── storage.py # multi-entity JSON storage │ ├── settings.py # local configuration & key management (AES) │ ├── sse.py # SSE tool │ ├── providers/ # multi-protocol LLM routing (core) │ │ ├── base.py # BaseProvider abstract │ │ ├── openai_provider.py │ │ ├── deepseek_provider.py │ │ ├── anthropic_provider.py │ │ ├── google_provider.py │ │ ├── agui_provider.py │ │ └── registry.py │ ├── paper/ # arXiv/PDF/BibTeX tool (Phase 2) │ └── prompts/ # various system prompt templates │ ├── routes/ # Flask Blueprint │ ├── items.py # knowledge entry CRUD │ ├── chat.py # multi-session chat (SSE) │ ├── providers.py # connection management │ ├── ingest.py # document import (Phase 2) │ └── settings.py # UI preferences │ ├── knowledge-base/ # user data (generated at runtime, .gitignore) ├── settings/ # connection configuration & preferences (generated at runtime, .gitignore) └── public/ # front-end static resources ├── index.html ├── styles/ └── js/ ``` ## Multi-Protocol Provider Design All providers implement a unified streaming interface, upper-layer routing and front-end completely unaware of differences: ```python class BaseProvider: def stream_chat(self, *, messages, model, temperature, system=None, max_tokens=None, extra=None): """yield {'type': 'text'|'thinking'|'error'|'done', 'delta': str, 'meta': {}}""" def list_models(self) -> list[dict]: ... def test_connection(self) -> dict: ... ``` Want to add a new provider? Three steps: 1. Create `xxx_provider.py` in `lib/providers/`, inherit `BaseProvider` 2. Register type in `lib/providers/registry.py` 3. Front-end `settings/providers.json` can save connection configuration of this type ## Security and Privacy - All API Keys are encrypted with **AES-256 + machine fingerprint** and stored in `settings/providers.json` - All data is stored locally, **not uploaded to cloud** - `.gitignore` has excluded `knowledge-base/`, `settings/`, `.venv/` and other directories ## License Personal Use — by BlackTea-c