Skip to content

GyaneshSamanta/Evols

Repository files navigation

Evols Hero

Evols

Your team's AI brain — knowledge, context, and coordination in one place.

Python FastAPI Next.js Postgres Last commit Status


About — the 5 Ws

What. Evols is a multi-tenant AI collaboration layer for product teams. It wraps an FastAPI backend, a Next.js workbench, a LightRAG knowledge graph, an OIDC bridge, and an MCP endpoint into a single platform that lets every teammate — and every AI tool they use — operate from the same shared context instead of rebuilding it from scratch.

Who. Built by Gyanesh Samanta and Akshay (Evols-AI co-founders). This repo is GyaneshSamanta's working fork of Evols-AI/Evols.

When. Active development from March 2026 → April 2026, with v1.0 (knowledge graph + clustering + personas), v1.1 (skills system + work context + dedup + retention), and v1.2 (Team Knowledge Graph + Claude Code/Zed plugins + AI Workbench) all shipped in that window.

Where. Built for product organizations who want their PMs, engineers, and AI assistants to stop re-deriving the same context every Monday morning. Deployed on Google Cloud Run with an Nginx reverse proxy gating a single public surface.

Why. Modern PM teams sit on a mountain of feedback, transcripts, decisions, and tribal knowledge — but the AI tools they reach for (Claude Code, Zed, ChatGPT) have no idea any of it exists. Evols is the brain those tools plug into, so the answer to "have we tried this before?" stops being "ask Slack."


The Story

It started with a familiar product-management tax: a PM uploading the same CSV of customer feedback into the third tool that month, then re-typing the same persona context into the fourth AI prompt of the day. Every tool was smart in isolation. The team, collectively, kept forgetting things.

The first version was a feedback-clustering pipeline — semantic embeddings, HDBSCAN, and an LLM theme generator that turned thousands of raw quotes into a handful of themes with urgency and impact scores. That worked, so we kept pulling at the thread. Themes earned Initiatives (the strategic response), and Initiatives earned Projects scored by a RICE formula: Priority = (Reach × PersonaWeight × Confidence) / Effort. Big work became 🏔️ Boulders; quick wins became ⚡ Pebbles. Suddenly there was a traceable line from a customer quote to a sprint ticket.

The second leap was admitting that the most valuable knowledge in an org doesn't live in CSVs — it lives in Claude Code transcripts, Zed sessions, and the Slack threads where someone said "yeah we tried that in Q1." So we built the Team Knowledge Graph: a LightRAG-backed store with a Streamable-HTTP MCP endpoint, a Claude Code plugin (SessionStart hook loads context, Stop hook auto-summarizes the session via Haiku), and a Zed extension that injects PM skills directly into every new AI thread. Redundancy detection became a one-line check: "someone already solved this two weeks ago, here's the entry."

The third leap was the AI Workbench — a LibreChat fork bolted to Evols via an OIDC provider bridge and a BYOK LLM proxy, so users SSO once, bring their own keys (AES-encrypted per tenant), and chat with Claude/GPT/Gemini/Ollama against their team's actual knowledge graph. Today the platform spans 80+ AI skills, 20+ ingestion source types, and three developer surfaces (web, Claude Code, Zed) — all backed by one shared FastAPI backend with Celery workers, pgvector, and a Bayesian outcome-learning loop that adjusts priorities based on what actually shipped.


Gallery

Collaboration Project Stages Features Overview Innovation Authentication Book Demo

Tech Stack

Backend

  • Python 3.13 · FastAPI 0.109 · SQLAlchemy 2.0 (async) · Alembic
  • PostgreSQL 15+ with pgvector · LightRAG (self-hosted graph RAG)
  • Celery + Redis for background jobs · psycopg3 · asyncpg
  • AWS Bedrock / OpenAI / Azure OpenAI / Anthropic (BYOK, AES-256-GCM encrypted per tenant)
  • sentence-transformers, scikit-learn, HDBSCAN for clustering
  • Tavily AI (primary) + Serper (fallback) for web search
  • JWT, OAuth2, OIDC provider bridge for SSO

Frontend

  • React 18 + Next.js 14 (App Router)
  • Tailwind CSS, shadcn/ui, Radix UI primitives
  • Zustand + TanStack Query (React Query v5)
  • D3.js, Recharts for visualization

AI Workbench

  • LibreChat fork (MIT) deployed to Cloud Run, MongoDB Atlas M0, gated by OIDC + LLM proxy

DevOps

  • Docker · Docker Compose · Google Cloud Run · Nginx (single public ingress)

Repository Structure

Evols/
├── backend/             # FastAPI service — APIs, models, services, Celery workers
│   ├── app/
│   │   ├── api/         # /api/v1/team-knowledge, /mcp, /oidc, /llm-proxy, ...
│   │   ├── core/        # auth, config, encryption
│   │   ├── models/      # SQLAlchemy ORM
│   │   ├── schemas/     # Pydantic request/response
│   │   ├── services/    # AI, clustering, persona, knowledge, RICE
│   │   └── workers/     # Celery tasks
│   └── requirements.txt
├── frontend/            # Next.js 14 + Tailwind + shadcn workbench UI
├── alembic/             # Database migrations
├── docker/              # backend/frontend Dockerfiles + docker-compose
├── deployment/          # Cloud Run + Nginx config
├── docs/                # CSV upload, troubleshooting, three-tier architecture
├── librechat.yaml       # AI Workbench wiring (LLM endpoints, MCP, auth)
├── SETUP_GUIDE.md       # End-to-end setup
└── CHANGELOG.md         # Version history (v1.0 → v1.2)

Getting Started

Prerequisites

  • Python 3.13+
  • Node.js 18+
  • PostgreSQL 15+ (with pgvector)
  • Docker & Docker Compose

Quick start (Docker)

git clone https://github.com/GyaneshSamanta/Evols.git
cd Evols

cp backend/.env.example backend/.env       # fill in DB + secrets
cp frontend/.env.example frontend/.env

docker-compose -f docker/docker-compose.yml up -d

Then visit:

Local development

# Backend
cd backend
python -m venv venv && source venv/bin/activate     # Windows: venv\Scripts\activate
pip install -r requirements.txt
uvicorn app.main:app --reload

# Frontend
cd frontend
npm install
npm run dev

After login, head to Settings → LLM Settings and add an OpenAI / Anthropic / Azure / AWS Bedrock key — AI features (skills, personas, themes) require BYOK credentials. See SETUP_GUIDE.md for the full walkthrough.


Contributing

# 1. Fork the repo on GitHub
git clone https://github.com/<you>/Evols.git
cd Evols

# 2. Branch, commit, push
git checkout -b feat/your-feature
git commit -m "feat: short description"
git push origin feat/your-feature

# 3. Open a PR against main

See CONTRIBUTING.md for the full workflow.


License

Copyright © 2026 Evols. All rights reserved.


Credits

  • Gyanesh Samanta — Co-founder, product, full-stack (@GyaneshSamanta)
  • Akshay — Co-founder, engineering

Documentation: docs.evols.ai · Email: support@evols.ai

About

Your team's AI brain — shared knowledge graph, MCP-connected workbench, and AI-powered product OS for PM teams.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors