AIHostBox
Client area Deploy an app
All 57 applications
CategoriesAI assistants & chatLLM runtimesAgents & RAGGateways & LLM opsVector & dataAutomationSpeech, image & videoSearch & web agentsSupport & CXML platformMCP servers
SolutionsPrivate team chatUnlimited automationCompany knowledge AIAI gateway & observability
More PricingGuides ComparisonsToolsPrivate LLM hosting LicensingSecurity SupportClient area

Letta hosting

Stateful agents with long-term memory that survives the context window.

Agents & RAGApache-2.0upstream discontinuedfrom $11.99/mo
In short

Letta hosting at AIHostBox means a dedicated, single-tenant Letta instance on our infrastructure: we install it, pin it to a reviewed release, put it behind TLS on your domain, back it up nightly and patch everything around it. Plans start at $11.99 per month and the admin account, the data and any API keys are yours.

No longer maintained upstream

Letta no longer maintains or supports the letta/letta Docker image as a product and directs new builds to the Letta App Server.

We still host it for teams who depend on it: we keep the instance running, backed up and behind TLS, and patch the operating system and the services around it. We cannot fix bugs or vulnerabilities inside Letta itself, and there will be no feature updates. Order it only if you accept that; if you are starting fresh, the alternatives below are the safer choice.

Persistent memoryAgent serverTool callingMulti-agentADE

Letta is the agent server that grew out of the MemGPT research project at UC Berkeley, now maintained by the company of the same name. Where most agent frameworks rebuild context on every request, Letta stores each agent as a persistent object: editable memory blocks that stay in the prompt, an archival memory searched by embeddings, a recall log of past messages, and the tools the agent may call. The agent itself decides what to write to memory and what to page back in.

Because that state lives in a database, a Letta agent is only as durable as the server holding it. Run it on a laptop and the agent forgets nothing but is unreachable whenever the machine sleeps. On a server it has a stable REST API that your application, a chat bot or a scheduled job can reach at any hour, with the memory store backed up nightly instead of sitting in a local Docker volume.

Letta hosting plans

PlusRecommended minimum
4 GB RAM · 2 vCPU · NVMe
$11.99/mo
Order
ProTeams & heavier load
8 GB RAM · 3 vCPU · NVMe
$18.99/mo
Order

Every plan is a dedicated single-tenant instance with NVMe storage, nightly off-box backups, TLS and managed patching. Quarterly, semi-annual and annual terms are discounted at checkout. Model tokens are not included — you bring your own key.

How we run Letta

The letta/letta server container exposing the REST API on port 8283, backed by Postgres with pgvector for agent state and archival memory.

What we configure for you

  • SECURE=true with a random LETTA_SERVER_PASSWORD
  • OPENAI_API_KEY, ANTHROPIC_API_KEY or OLLAMA_BASE_URL for your chosen providers
  • Postgres with pgvector on a persistent volume (or LETTA_PG_URI to an external database)
  • TLS on your subdomain so the Agent Development Environment can connect
  • E2B_API_KEY if you want custom tools executed in E2B sandboxes

Which plan fits

Plus (4 GB) runs the server and database for a developer or a small product with a modest number of agents. Pro (8 GB) suits many agents with large archival memories, heavier concurrent traffic, or a local Ollama embedding model alongside.

The honest limits

Letta now describes the Docker server image as no longer an actively supported product surface and points new local builds at its App Server, so check the current state with us before building a long-term product on it. Memory quality depends heavily on the model; small CPU-hosted models are unreliable at Letta’s tool-driven memory edits.

What people run Letta for

Assistants that remember

Continuity across weeks of conversation, not one session.

Long-running agents

State that outlives a single request.

Per-customer agents

One agent per user, each with its own memory blocks, served from a single API.

Licence and version policy

ItemPosition
Upstream licenceApache-2.0
Sold as managed hostingYes
TenancySingle tenant — one customer per instance
PatchingCritical advisories within 24 hours, high severity within 72

Our full position on licences, including the applications we decline to sell, is on the licensing page.

Reading on Letta

ComparisonOpenClaw alternativesGuide · AgentsAI agent hosting: where an agent should actually run

Works well with Letta

Ollama logo

Ollama

Your private LLM server — pull a model, get an API, keep every token on your own box.

from $16.99/moDetails →Hermes logo

Hermes

Nous Research’s Hermes models, served privately behind your own API.

from $16.99/moDetails →LiteLLM logo

LiteLLM

One OpenAI-compatible endpoint in front of every model provider you use.

from $5.99/moDetails →Supergateway logo

Supergateway

Put a stdio MCP server on the network without rewriting it.

from $5.99/moDetails →

More Agents & RAG we host

Dify logo

Dify

A visual studio for building LLM apps, agents and RAG pipelines.

from $18.99/moDetails →Flowise logo

Flowise

Drag-and-drop LLM flows, agents and chatbots you can embed anywhere.

from $11.99/moDetails →Langflow logo

Langflow

A visual builder for LangChain flows and multi-agent systems.

from $11.99/moDetails →AnythingLLM logo

AnythingLLM

A private document workspace — drop files in, ask questions, get cited answers.

from $18.99/moDetails →
Questions

Letta hosting: your questions

Which models can Letta agents use?
Letta reads provider keys such as OPENAI_API_KEY and ANTHROPIC_API_KEY from its environment, and OLLAMA_BASE_URL for local models. Agents also need an embedding model for archival memory, which must be chosen explicitly when running the Docker server. Weak models struggle with Letta’s memory-editing tools, so small local models are best kept for testing.
How is the server protected?
We start it with SECURE=true and a generated LETTA_SERVER_PASSWORD, so every API call must carry that password in its Authorization header. It is served over TLS on your own domain, which is also what the browser-based Agent Development Environment needs before it will connect to a remote server.
Where is agent memory stored and backed up?
Agents, memory blocks, message history and archival passages all live in Postgres with the pgvector extension. That database is included in the nightly off-box backup, so restoring it restores every agent exactly as it was, including what it had chosen to remember.
Can I bring agents over from my own Letta install?
Yes. If you ran the letta/letta image with its bundled database, we can restore a Postgres dump into your instance. Agents can also be exported and imported individually through Letta’s agent file format if you only want to move a few. Tell us which version you ran so migrations line up.
Does Letta support MCP tools?
Yes. Letta can connect to MCP servers and attach their tools to agents alongside its built-in memory tools. A remote MCP server, for example one bridged with Supergateway, is reachable over the network, while stdio servers need to run where the Letta server can start them.

Run Letta without running a server

A dedicated Letta instance, deployed and maintained by us. Month to month, cancel whenever.

Browse the catalogue Talk to a human