Letta agent platform for AI assistants that keep memory across sessions
Letta is an agent platform for AI assistants that keep memory across sessions. See how MILabs sets it up for small businesses.
Updated 2026-02-11
What Letta is
Letta is a platform for building AI agents that keep their own memory over time. It grew out of the MemGPT research project and is maintained by the company Letta. Where a plain chat session forgets earlier conversations once it closes, a Letta agent is designed to store what it learns and draw on it in later sessions. Developers can run the open-source server on their own infrastructure or use the hosted service, and the agents are reached through an API and related tooling. Businesses look at Letta when they want an assistant that remembers a customer's history, a staff member's preferences or the details of an ongoing project, rather than starting from zero each time. It is a developer-oriented tool, so most of the value comes from how the agent is configured, what memory it is given and which systems it may read and write. For a small business, Letta is usually one component inside a larger workflow rather than a finished product on its own.
Letta in plain language
A large language model on its own answers each request without any recollection of earlier requests, so a business that wants continuity has to supply the context again and again. Agent platforms add a layer around the model that holds information, decides what to retrieve and calls outside tools to act on the result. Memory is the feature that separates an agent from a single chat window. Some memory lasts only as long as one task, while other memory is kept so the agent can recall facts later on.
Letta's documentation describes memory as a set of editable blocks that the agent can read and update during a conversation, rather than a transcript that simply grows. Because the agent writes to its own memory, the contents become something a business has to look after. Owners should expect to decide what is worth keeping, how long it stays and which staff can view or correct it. A useful comparison is a shared client file, which is only as reliable as the habits of the people who keep it up to date.
What businesses use it for
- Keeping a running record of each client's history so follow-up messages reflect earlier conversations.
- Giving staff a knowledge assistant that recalls how the team prefers reports and briefings to be set out.
- Tracking the status of a long-running project across many separate sessions.
- Holding onboarding context for new hires as they settle into their roles.
- Supporting an intake assistant that recognises a returning customer's preferences.
- Maintaining a memory of supplier terms and past orders to inform purchasing decisions.
Who it suits
- Teams that want an assistant to keep context across ongoing client work.
- Businesses with technical staff who can run and monitor a self-hosted service.
- Owners who want agent memory kept within infrastructure they control.
How MILabs sets it up
Setup starts with a conversation with the owner or team lead about what the assistant should remember. We help choose between running Letta on infrastructure the business controls or a hosted option, based on where customer data should sit. Next we connect the systems the business already uses, such as the inbox, CRM or booking calendar, so the agent reads and writes only the records it needs. Permissions are set for which memories each staff role can view, which fields the agent can change, and what it must hand back to a person. Finally we agree who looks after the agent in routine use, who reviews its memory for errors and who is contacted when something misbehaves. Delivery is remote outside Sydney, and the same steps apply wherever the business is located.
Builds it fits
Industries it fits
Letta in practice
Typically, the agent sits behind a single channel such as a website chat window or a shared inbox, and each conversation is tied to a stored record for the person or account involved. When a returning enquiry arrives, the agent retrieves the relevant memory before it replies, so the message can refer to an earlier quote or a stated preference. The business keeps its source records in the systems it already uses, and the agent's memory is a derived layer that can be reviewed and reset.
Staff usually interact with the agent differently from customers. A team member might ask it to summarise a project, recall how a supplier prefers orders placed, or draft a handover note. Each of these requests depends on memory that was written deliberately, which is why the build begins with a list of what the agent is permitted to store.
Over time, the work shifts from building to upkeep. Someone reviews a sample of remembered entries on a regular schedule and corrects anything out of date. A short written note describing what the agent remembers, who may change it and how a person can override a reply keeps the arrangement clear to the next person who inherits it.
Frequently asked questions
- What do businesses use Letta for?
- Businesses usually use Letta when an assistant needs to remember earlier interactions. Common examples include keeping a client's history so follow-up replies stay consistent, giving staff a knowledge assistant that recalls how the team works, and carrying context through a long onboarding or project. The value depends on the agent's memory being scoped carefully, so it holds only what the business has agreed to keep.
- Does Letta need technical skills to run?
- Yes, in most cases. Letta is a developer-oriented platform, and running the server, managing its database and configuring agents typically requires someone comfortable with APIs and command-line tools. Businesses without that skill can have MILabs handle the build and hand over a documented setup, and agree on who monitors the agent afterwards. The learning curve is one reason we recommend deciding the hosting option early.
- Can Letta connect to the software I already use?
- Yes. Letta exposes its agents through an API, so it can be connected to other systems by code or through automation tools. Whether a given product can be linked depends on that product's own interface, and we confirm this during setup rather than assume it. Typical connections for a small business include the inbox, a CRM, a booking calendar or a shared document store. We limit each connection to the fields the agent actually needs.
- How does Letta compare with other tools for the same job?
- Letta focuses on agents with persistent memory, unlike tools built mainly for chat, visual workflows or document retrieval. Workflow tools such as n8n or Zapier are built mainly to move data between apps, so agent memory has to be designed around them. Frameworks such as LangChain or CrewAI offer agent building blocks that can include memory. The right choice depends on the job, the team's skills and where the data must live.
- How do I get my team to use Letta?
- Start with a single, well-defined task that the team already finds tedious, and let staff see how the assistant answers before expanding it. Agree who can correct the agent's memory, because staff trust grows when mistakes are fixed quickly. Short internal walkthroughs, written guidance and a named contact for questions usually matter more than the tool itself. Any business that wants to discuss a first project can get in touch through the enquiry form.
