⭐ Star MemGPT on GitHub

Frequently asked questions (FAQ)

Check frequently asked questions

📘

Open / local LLM FAQ

Questions specific to running your own open / local LLMs with MemGPT can be found here.

MemGPT CLI

How can I use MemGPT to chat with my docs?

Check out our chat with your docs example to get started.

How do I save a chat and continue it later?

When you want to end a chat, run /exit, and MemGPT will save your current chat with your agent (make a note of the agent name, e.g. agent_N). Later, when you want to start a chat with that same agent, you can run memgpt run --agent <NAME>.

How do I implement MemGPT for multiple users?

The REST API for MemGPT is flexible and leverages PostgreSQL DB or SQLite for its backend. To implement a multi-user setup, first determine the user_id (either create a UUID or use the user_id from your own database). Then create an agent, and finally use the agent_id and user_id to post a message or run a command. Internally the following occurs:

  • a user creates an agent
  • that agent is "owned" by a user
  • when the user sends the agent a message, that's stored in a message collection (messages are indexed by user and agent ids)
  • on the higher-level agents side (not talking about db implementation details), the agent can only see a few messages at a time, but has access to all the messages ever sent between it and the user via the recall memory search functions
  • the database is multi-user, and the REST endpoints function in a way where user data is not shared

My MemGPT agent is stuck "Thinking..." on the first message?

MemGPT has an extra verification procedure on the very first message to check that in the first message (1) the agent is sending a message to the user, and (2) that the agent is using internal monologue. This verification is meant to avoid the scenario where a bad initial agent message "poisons" the rest of a conversation. For example, a message missing internal monologue might cause all future messages to also omit internal monologue.

If the LLM/model you're using for MemGPT is consistently failing the first message verification, it will appear as a long "Thinking..." loop on the first message. "Weaker" models such as gpt-3.5-turbo can frequently fail first message verification because they do not properly use the send_message function and instead put the message inside the internal monologue. Better models such as gpt-4 and gpt-4-turbo, as well as open models like dolphin-2.2.1 and openhermes-2.5 should not have this problem.

You can disable first message verification by passing the --no-verify flag to memgpt run (do memgpt run --no-verify instead of memgpt run). Passing the additional --debug flag (memgpt run --no-verify --debug) can help you further identify any other issues on first messages that can cause long "Thinking..." loops, such as rate limiting.

What are personas and how they relate to agents and humans?

MemGPT has two core components: agents and humans. Each human contains information about the user that is continously updated as MemGPT learns more about that user. Agents are what the human interacts with when they chat with MemGPT. Each agent can be customized through presets which are basically the configuration for an agent and includes the following componenets:

  • system prompt (you usually don't change this)
  • persona (personality of your bot and their initial memories)
  • human (description of yourself / user details)
  • functions (the functions the agent can call during convo)

I broke/corrupted my agent, how can I restore an earlier checkpoint?

MemGPT saves agent checkpoints (.json files) inside the ~/.memgpt/agents/YOUR_AGENT_NAME/agent_state directory (C:\Users\YourUsername\.memgpt\YOUR_AGENT_NAME\agent_state on Windows). By default, when you load an agent with memgpt run it will pull the latest checkpoint .json file to load (sorted by date).

If you would like to revert to an earlier checkpoint, if you remove or delete other checkpoint files such that the specific .json from the date you would like you use is the most recent checkpoint, then it should get automatically loaded by memgpt run. We recommend backing up your agent folder before attempting to delete or remove checkpoint files.

OpenAI-related

How do I get an OpenAI key?

To get an OpenAI key, visit https://platform.openai.com/, and make an account.

Then go to https://platform.openai.com/account/api-keys to create an API key. API keys start with sk-....

How can I get gpt-4 access?

https://help.openai.com/en/articles/7102672-how-can-i-access-gpt-4

I already pay for ChatGPT, is this the same as GPT API?

No, ChatGPT Plus is a separate product from the OpenAI API. Paying for ChatGPT Plus does not get you access to the OpenAI API, vice versa.

I don't want to use OpenAI, can I still use MemGPT?

Yes, you can run MemGPT with your own LLMs. See our section on local LLMs for information on how to set them up with MemGPT.

Updated 4 months ago