GeepSeek

Overview

Python Flask Ollama SQLite SSE Streaming Search Agent OpenAI API Jinja2 JavaScript

Fully Local — Runs entirely on your hardware via Ollama, no external API keys required

Search Agent — Autonomous web browsing agent retrieves real-time information

Persistent Memory — SQLite-backed conversation storage with session management

Streaming Responses — Real-time SSE chat with reasoning display and content blocks

Provider Agnostic — Any OpenAI-compatible endpoint can serve as the model backend

GeepSeek is a two-tier application designed for developers and privacy-conscious users who want to run AI models locally. A Flask web client serves the UI while a separate Flask API handles inference, search, and persistence — giving you a full ChatGPT-like experience without sending data to external servers.

The client serves HTML and static assets on its own port. The browser communicates with the API for sessions, history, and chat streaming. When a user sends a message, the client POSTs to the chat endpoint with the session ID, message content, and feature flags. If search is enabled, the search agent retrieves web content before the main model responds.

The server streams reasoning and content chunks as SSE events, displaying the model's thought process in real-time alongside final answers. Three SQLite databases store chat messages, session metadata, and optional UI state. Each chat session maps to a dedicated message table, keeping conversation data clean and isolated.

The architecture follows local-first principles — no external database required, the UI and API run as independent processes, and any OpenAI-compatible endpoint (including Ollama) can serve as the inference backend. This makes GeepSeek ideal for experimentation, development, and anyone who values running AI on their own terms.

Overview

Gallery