Get started
Introduction
What MenteDB is, how the one call per turn works, and where to start.
MenteDB gives your agent long term memory through one call, process_turn. Each turn it stores what was said, retrieves what is relevant, and hands back the memories to put in your next prompt. That loop is the whole integration.
Under it is a storage engine written in Rust, not a wrapper around a vector database. It combines vector search, a knowledge graph, facts that know when they stopped being true, and context assembly built for the way a model reads a prompt, so what your agent reads is clean, current and in the right order.
Choose a path
One call per turn
This is the whole loop on the cloud API. The embedded engine and the self hosted server take the same fields.
import os, requests
resp = requests.post(
"https://api.mentedb.com/v1/process_turn",
headers={"Authorization": f"Bearer {os.environ['MENTEDB_API_KEY']}"},
json={
"user_message": "I switched our side projects from Postgres to SQLite",
"assistant_response": "Noted.",
"turn_id": 0,
"session_id": "conv-7",
},
)
context = resp.json()["context"] # memories to put in the next promptExplore the docs
Core concepts
- Inside process_turn
- Memory types and the graph
- Context assembly
- Search and reranking
- Provenance
- and 6 more in the sidebar
API reference
Hosting
- MenteDB Cloud
- Self hosting
- Docker
- Databases and tokens
- AWS Bedrock
- and 2 more in the sidebar
mente (Spanish): mind, intellect.