← Back to catalog Letta · managed by DINAO

AI agents with real memory

The Letta platform for AI agents with persistent memory (ex-MemGPT), installed and maintained by DINAO. Your agents and their memories live on our French servers — in local mode, they never leave the premises.

View compatible offers →
Hosted in FranceSovereign in local modeLong-term memoryGDPR CompliantOfficial publisher image
Overview

What is Letta?

Letta (formerly MemGPT) is an open-source platform dedicated to stateful AI agents : assistants equipped with advanced and persistent memory, capable of learning and improving over time, whereas a standard agent forgets everything from one conversation to the next.

The platform exposes a Letta API to integrate agents into your applications, a CLI (Letta Code) for local execution, as well as skills and subagents mechanisms. It is model-agnostic : you can drive any LLM, hosted locally or in the cloud, using Python and TypeScript SDKs.

Technically, the Letta server relies on PostgreSQL with the pgvector extension to store agent memory durably. Deployed in containers by DINAO and connected to a local model, it becomes a sovereign agent foundation, hosted in France.

Compatible offers

Host Letta at DINAO

Resource tiers compatible with Letta prerequisites (minimum 2 vCPU / 2 Go / 10 Go). Hosted in France, fully managed.

Standard
2 vCPU · 4 Go · 40 Go
19,90 € /month excl. VAT
  • 2 dedicated vCPU
  • 4 Go RAM
  • 40 GB NVMe
  • Daily backups
  • Managed & monitored by DINAO
Order
★ Recommended for this app
Performance
4 vCPU · 8 Go · 80 Go
39,90 € /month excl. VAT
  • 4 dedicated vCPU
  • 8 Go RAM
  • 80 GB NVMe
  • Daily backups
  • Managed & monitored by DINAO
Order
Dédié
8 vCPU · 16 Go · 160 Go
79,90 € /month excl. VAT
  • 8 dedicated vCPU
  • 16 Go RAM
  • 160 GB NVMe
  • Daily backups
  • Managed & monitored by DINAO
Order

1 hidden tier(s) (insufficient resources for this app) : Découverte

🧠

This application uses AI

The container hosts the application, not the AI engine (which requires dedicated GPUs) : inference is handled externally, with your own provider key. Prioritize a sovereign engine — Mistral AI, NumSpot, Scaleway, or OVHcloud AI Endpoints (France, GDPR) ; an international provider (OpenAI, Anthropic…) only if a specific capability requires it. Inference subscriptions are not included in hosting.

Under the hood

Technical details

vCPU
2 vCPU
ideal : 4 vCPU
Memory
2 Go
ideal : 4 Go
Disk
10 Go
ideal : 20 Go
Image : letta/letta:latest Registry : docker.io Services : letta_db, letta_server Ports : 8283
FAQ

You might be wondering…

How does Letta differ from a simple chatbot?

Letta manages stateful agents : they have long-term memory and retain their context over time, instead of starting from scratch in each session. This is the legacy of the MemGPT project.

Is my agents' memory private?

Yes. In local model mode, agents, memories, and conversations remain on your instance hosted in France, with no transmission to third parties or use for training.

What language model does Letta use?

Letta is model-agnostic: it orchestrates agents but does not provide the model. You connect a local open-weight LLM (Ollama) for sovereignty, or an external API with your key.

How do I integrate Letta into my application?

Via the Letta API and Python / TypeScript SDKs, or the CLI for local agent execution. DINAO hosts the server and its PostgreSQL database (pgvector); you develop your agents.

Do I need a GPU?

Not necessarily. Small models (3–7B) run on CPU. For more powerful agents or high throughput, a GPU is recommended: offered as an option on Team and Enterprise tiers.