DEDUPED Hot AI repos. Real use cases. What people build. Get the Monday 5

How to build a private knowledge graph with local models with Graphiti

Run extraction and embeddings through Ollama's OpenAI-compatible endpoint so no data leaves your machine.

Short answer: Pull the models with ollama pull deepseek-r1:7b and ollama pull nomic-embed-text; then make sure Ollama is running with ollama serve; configure OpenAIGenericClient with base_url http://localhost:11434/v1 and an OpenAIEmbedder with embedding_dim 768, as in the README. All steps.

  • Research
  • Advanced
  • 2 hours
  • From the README
  • Claude Code
  • Codex
  • Cursor
  • Gemini CLI

Who it is forA privacy-minded developer who cannot send documents to a hosted model.

The problemThe default setup sends everything to OpenAI.

The outcomeGraphiti running against local deepseek-r1:7b and nomic-embed-text models via OpenAIGenericClient.

What you need first

  • Python 3.10 or higher; install with `pip install graphiti-core`
  • A graph database: Neo4j 5.26, FalkorDB 1.1.2, Amazon Neptune, or Kuzu (deprecated)
  • An OpenAI API key by default, or another provider such as Anthropic, Gemini, Groq or a local Ollama server

Steps

  1. Pull the models with ollama pull deepseek-r1:7b and ollama pull nomic-embed-text
  2. Make sure Ollama is running with ollama serve
  3. Configure OpenAIGenericClient with base_url http://localhost:11434/v1 and an OpenAIEmbedder with embedding_dim 768, as in the README
  4. Set structured_output_mode to json_object if your local model does not honor json_schema

Prompt for your agent

Using getzep/graphiti with a local Ollama server, configure OpenAIGenericClient and OpenAIEmbedder as the README shows, ingest <my notes folder>, and report how many entities were extracted and any extraction failures.

Caveat. Very small local models often emit JSON that does not match the schema; prefer the most capable model you can run.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for getzep/graphiti and "Build a private knowledge graph with local models"

Add Deduped to Claude Code once

claude mcp add --transport http deduped https://dedupedai.com/mcp

Cursor, ChatGPT, others

Share: X LinkedIn Hacker News Email Copy link

Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer