TL;DR: MemoryOS is 100% local by default. Your data never leaves your machine.
| File | Contents | Location |
|---|---|---|
memories.db |
All memory records, sessions, tags | ~/.memoryos/ |
chroma/ |
Vector embeddings (numbers, not text) | ~/.memoryos/ |
models/all-MiniLM-L6-v2/ |
Embedding model weights | ~/.memoryos/ |
server.pid |
Server process ID | ~/.memoryos/ |
.env |
Your configuration | ~/.memoryos/ |
- β Your chat content
- β Your memory data
- β Your usage patterns
- β Any analytics or telemetry
- β
Your API keys (stored only in
~/.memoryos/.env)
| Service | When Called | What's Sent | How to Disable |
|---|---|---|---|
| Ollama | Summarization | Memory content | Runs locally β no network |
| Groq API | Summarization fallback | Memory content (truncated to 2000 chars) | Remove GROQ_API_KEY from .env |
| OpenAI API | Summarization fallback | Memory content (truncated to 2000 chars) | Remove OPENAI_API_KEY from .env |
| OpenAI Embeddings | If EMBEDDING_PROVIDER=openai |
Memory content | Set EMBEDDING_PROVIDER=local |
| PyPI | Install only | Package metadata | One-time only |
| HuggingFace | First run only | Nothing (downloads model) | Pre-download model |
The extension:
- Only communicates with
http://localhost:8765 - Never sends data to any external server
- Never reads page content outside of the whitelisted AI sites
- Only captures conversation turns (user + assistant messages)
- Uses content hashing to avoid sending duplicates
# Forget a specific memory
memoryos forget <id>
# Delete ALL data
rm -rf ~/.memoryos/That's it. No servers to contact, no account to delete.
Since all data is local and you control it entirely:
- Right to access:
memoryos export --format json - Right to erasure:
rm -rf ~/.memoryos/ - Right to portability:
memoryos export --format jsonor--format markdown - Data controller: You.