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

How to give an agent memory that tracks how facts change with Graphiti

Store each conversation turn or record as an episode so the graph keeps what is true now and what was true before.

Short answer: Install the library with pip install graphiti-core and set OPENAI_API_KEY in your environment; then start a graph database with docker compose up for Neo4j, or docker compose --profile falkordb up for FalkorDB; follow the Quickstart Example to connect, build the indices and constraints, and add text and JSON episodes. All steps.

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

privacy. Episode content is sent to the configured LLM provider for extraction; use a local model for sensitive data.

cost. Ingestion calls the LLM for every episode; the README sets SEMAPHORE_LIMIT to 10 by default to avoid 429 rate limit errors.

Who it is forA developer whose assistant forgets or contradicts user preferences that change over time.

The problemFlat document chunks cannot tell an old fact from its replacement.

The outcomeA context graph where superseded facts are invalidated, not deleted, and every fact traces back to its episode.

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. Install the library with pip install graphiti-core and set OPENAI_API_KEY in your environment
  2. Start a graph database with docker compose up for Neo4j, or docker compose --profile falkordb up for FalkorDB
  3. Follow the Quickstart Example to connect, build the indices and constraints, and add text and JSON episodes
  4. Query with hybrid search for relationships and rerank results by graph distance

Prompt for your agent

Using getzep/graphiti, install it with `pip install graphiti-core`, start Neo4j with `docker compose up`, ingest <my conversation log> as episodes, and show me one fact that was invalidated by a later episode.

Caveat. Graphiti works best with LLM services that support structured output; smaller models may cause ingestion failures.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for getzep/graphiti and "Give an agent memory that tracks how facts change"

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