KnowBear Knowledge Engine

AI-powered layered explanations for any topic. Built with FastAPI (Python) and React (TypeScript).

Status

This repository is the deprecated KnowBear v1 codebase, kept for historical reference.

• Active repository (v2): https://github.com/voidcommit-afk/know-bear
• This repo (v1): legacy snapshot, no active feature development

Features

• Layered Explanations: ELI5 to ELI15, meme-style, technical deep dives
• Mode-Based Routing: fast uses a single low-latency model, ensemble runs multi-model generation with judge-based selection
• Multi-Model Ensemble: Parallel generation with judge-based voting
• Redis Caching: Smart caching for fast repeat queries
• Export Options: Download as .txt, .json, or .pdf
• Dark Theme UI: Minimalist, space-themed design

Model Routing (Actual v1)

Query Modes (/api/query)

• fast mode:

  • Uses one Groq model: llama-3.1-8b-instant
  • Lower latency path (no judge)

• ensemble mode:

  • Runs these Groq models in parallel:
    • llama-3.1-8b-instant
    • llama-3.3-70b-versatile
    • llama-3.1-70b-versatile
    • deepseek-r1-distill-llama-70b
    • mixtral-8x7b-32768
  • Uses llama-3.3-70b-versatile as a judge to select the best response
  • Server enforces premium gating for ensemble

Fallback Behavior

• If Groq fails, the backend fallback chain is:
  1. Hugging Face Inference API (microsoft/Phi-3-mini-4k-instruct) when HF_TOKEN is configured
  2. Google Gemini (gemini-2.0-flash) when GEMINI_API_KEY is configured

Streaming Notes (/api/query/stream)

• Streaming path uses llama-3.1-8b-instant in current implementation.
• Response chunks are adaptively buffered/flushed for smoother UX, with truncation signaling when token limits are hit.

Architecture

KnowBear/
├── api/               # FastAPI backend (Serverless compatible)
│   ├── main.py        # App entry point
│   ├── routers/       # API endpoints
│   └── services/      # Business logic (Routing, Inference, Cache)
├── src/               # React frontend
│   ├── components/    # Reusable UI components
│   ├── pages/         # Page layouts
│   └── hooks/         # Custom React hooks
├── public/            # Static assets
└── vercel.json        # Deployment configuration

API Endpoints

Endpoint	Method	Description
/api/pinned	GET	Curated popular topics
/api/query	POST	Generate layered explanations
/api/export	POST	Export results as file
/api/health	GET	System health check (Redis/Dependencies)

Tech Stack

• Frontend: React, Vite, TailwindCSS, Framer Motion
• Backend: FastAPI, Pydantic, Structlog
• AI/LLM: Groq (Llama, DeepSeek, Mixtral), Google Gemini, Hugging Face Inference API
• Database/Cache: Redis (Upstash/Cloud), Supabase (Auth)