> ## Documentation Index
> Fetch the complete documentation index at: https://docs.presenton.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Presentation memory

> Configure, persist, size, and troubleshoot self-hosted presentation memory.

Presentation memory helps AI-assisted editing retain relevant context across changes to the same presentation. Self-hosted Presenton uses Mem0 OSS with local vector and history storage under `/app_data` by default.

## Default configuration

```dotenv theme={null}
MEM0_ENABLED=true
MEM0_DIR=/app_data/mem0
MEM0_EMBEDDER_PROVIDER=fastembed
MEM0_EMBEDDER_MODEL=BAAI/bge-small-en-v1.5
MEM0_EMBEDDING_DIMS=384
MEM0_SPACY_MODEL=en_core_web_sm
MEM0_REQUIRE_SPACY_MODEL=true
```

The official container includes the required spaCy model. The embedding model may be downloaded on first use, so restricted or offline deployments should pre-populate the required cache.

## Memory model

Mem0 also needs an OpenAI-compatible text endpoint for extracting useful memories:

```dotenv theme={null}
MEM0_LLM_MODEL=llama3.1:latest
MEM0_LLM_API_KEY=ollama
MEM0_LLM_BASE_URL=http://host.docker.internal:11434
```

The memory model can differ from the model that creates presentations. Confirm that a container can reach the configured URL; `localhost` inside Docker does not refer to the host.

## Persistence and privacy

* Keep `MEM0_DIR` inside the persistent `/app_data` mount.
* Include memory data in backups and protect it like presentation content.
* Hosted memory-model endpoints receive the context sent for memory extraction.
* Set `MEM0_ENABLED=false` when iterative memory is unnecessary or the workflow must avoid that processing.

## Troubleshoot

Check application logs for missing spaCy or embedding models, unreachable endpoints, invalid dimensions, and filesystem permission errors. After changing the embedding model or dimensions, use a clean compatible collection rather than mixing incompatible vectors.