Hub/Docs

Overview

The knowledge graph for the Latent Space community — two repos, one database, four surfaces.

The System

Two repos, one knowledge graph.

RepoWhat it isDeployed on
latent-space-hubKnowledge graph — Next.js web app, ingestion pipeline, MCP server, evalsVercel
latent-space-botsDiscord bot (Slop) — agentic MCP tool-calling, member memoryRailway

They share one Turso database. The hub writes to it (ingestion, web UI, API). The bot reads from it (MCP tool calls) and writes member data back.

How It Works

Content sources             AI enrichment                Surfaces
──────────────────         ──────────────────           ──────────────────
YouTube transcripts    →   Embeddings (1536d)       →   Web App (Next.js)
Substack articles      →   Chunk splitting          →   MCP Server (agents)
AINews (smol.ai)       →   Entity extraction        →   Discord Bot (Slop)
LatentSpaceTV          →   Auto-edge creation       →   Announcements webhook
  1. Content goes in — auto-ingestion pipeline polls RSS feeds hourly, extracts from YouTube and Substack
  2. AI enriches it — chunks are embedded, entities extracted, edges created automatically
  3. Humans and agents explore it — via the web UI, MCP tools, or Discord
  4. The graph grows — each new piece of content connects to existing knowledge

Architecture

Four cloud services, no self-hosted infrastructure.

ServiceRole
VercelWeb app (read-only) + hourly ingestion cron + Discord announcements
TursoCloud SQLite — single shared database for web app, bot, and MCP
RailwaySlop bot — always-on Discord gateway with agentic MCP tool-calling
OpenRouterLLM routing for Slop (currently Claude Sonnet 4.6, swappable)

The Graph

MetricCount
Total nodes~3,900
Content nodes~515 (podcasts, articles, AINews, workshops, paper clubs, builders clubs)
People & entity nodes~3,360 (guests, organizations, topics)
Edges~7,500
Chunks with embeddings~35,800
CoverageJune 2023 → present, continuously updated

Tech Stack

LayerTechnology
FrameworkNext.js 15 + TypeScript + Tailwind CSS
DatabaseTurso (cloud SQLite via @libsql/client)
SearchTurso native vector search (F32_BLOB + vector_top_k) + FTS5 + hybrid RRF
EmbeddingsOpenAI text-embedding-3-small (1536d)
AIAnthropic Claude (entity extraction) + OpenAI (embeddings)
MCPModel Context Protocol server — npx latent-space-hub-mcp
BotDiscord.js + OpenRouter → Claude Sonnet 4.6

Built on RA-H, a local-first personal knowledge graph.