Player not loading? Watch on YouTube
This sponsored tutorial adds Mem0 memory to two agents and tests whether stored dietary facts survive a process restart. The OpenAI Agents SDK build exposes save and search functions as tools, with instructions to retrieve relevant memories before answering. It requires OpenAI and Mem0 keys. The speaker reports that a fresh run retrieves the facts and uses them to suggest dinner.
Hermes, the terminal agent from Nous Research, uses Mem0 as a pluggable provider alongside its own file notes. The walkthrough describes background memory extraction and cached searches for the next turn. The model can also search, add facts verbatim, update them by ID, and delete them. After five consecutive provider failures, Hermes pauses memory calls for two minutes while continuing to respond.
The speaker's inspection describes structured facts, linked updates rather than silent overwrites, temporary numeric IDs for model input, and dates grounded before storage. The metadata identifies the inspected package as mem0ai 2.0.18. For deployment, the tutorial contrasts managed cloud service with an open source, self-hosted configuration using your own model, embedder, and vector store; the speaker says that configuration keeps data on the machine. Practical caveats include a large extraction prompt on writes, cleanup policies for changing facts, and different user-filter parameters for add and search.
