Try out our MemGPT chatbot on Discord!
โญ NEW: You can now run MemGPT with local LLMs and AutoGen! โญ
Join Discord and message the MemGPT bot (in the #memgpt
channel). Then run the following commands (messaged to "MemGPT Bot"):
/profile
(to create your profile)/key
(to enter your OpenAI key)/create
(to create a MemGPT chatbot)
Make sure your privacy settings on this server are open so that MemGPT Bot can DM you:
MemGPT โ Privacy Settings โ Direct Messages set to ON
You can see the full list of available commands when you enter /
into the message box.
Memory-GPT (or MemGPT in short) is a system that intelligently manages different memory tiers in LLMs in order to effectively provide extended context within the LLM's limited context window. For example, MemGPT knows when to push critical information to a vector database and when to retrieve it later in the chat, enabling perpetual conversations. Learn more about MemGPT in our paper.
Install MemGPT:
pip install pymemgpt
Add your OpenAI API key to your environment:
export OPENAI_API_KEY=YOUR_API_KEY # on Linux/Mac
set OPENAI_API_KEY=YOUR_API_KEY # on Windows
$Env:OPENAI_API_KEY = "YOUR_API_KEY" # on Windows (PowerShell)
Configure default setting for MemGPT by running:
memgpt configure
Now, you can run MemGPT with:
memgpt run
The run
command supports the following optional flags (if set, will override config defaults):
--agent
: (str) Name of agent to create or to resume chatting with.--human
: (str) Name of the human to run the agent with.--persona
: (str) Name of agent persona to use.--model
: (str) LLM model to run [gpt-4, gpt-3.5].--preset
: (str) MemGPT preset to run agent with.--first
: (str) Allow user to sent the first message.--debug
: (bool) Show debug logs (default=False)--no-verify
: (bool) Bypass message verification (default=False)--yes
/-y
: (bool) Skip confirmation prompt and use defaults (default=False)
You can run the following commands in the MemGPT CLI prompt:
/exit
: Exit the CLI/attach
: Attach a loaded data source to the agent/save
: Save a checkpoint of the current agent/conversation state/dump
: View the current message log (see the contents of main context)/memory
: Print the current contents of agent memory/pop
: Undo the last message in the conversation/heartbeat
: Send a heartbeat system message to the agent/memorywarning
: Send a memory warning system message to the agent
Once you exit the CLI with /exit
, you can resume chatting with the same agent by specifying the agent name in memgpt run --agent <NAME>
.
You can add new human or persona definitions either by providing a file (using the -f
flag) or text (using the --text
flag).
# add a human
memgpt add human [-f <FILENAME>] [--text <TEXT>]
# add a persona
memgpt add persona [-f <FILENAME>] [--text <TEXT>]
You can view available persona and human files with the following command:
memgpt list [human/persona]
MemGPT supports pre-loading data into archival memory. You can attach data to your agent (which will place the data in your agent's archival memory) in two ways:
- Run `memgpt attach --agent --data-source
- While chatting with the agent, enter the
/attach
command and select the data source.
We currently support loading from a directory and database dumps. We highly encourage contributions for new data sources, which can be added as a new CLI data load command.
Loading from a directory:
# loading a directory
memgpt load directory --name <NAME> \
[--input-dir <DIRECTORY>] [--input-files <FILE1> <FILE2>...] [--recursive]
Loading from a database dump:
memgpt load database --name <NAME> \
--query <QUERY> \ # Query to run on database to get data
--dump-path <PATH> \ # Path to dump file
--scheme <SCHEME> \ # Database scheme
--host <HOST> \ # Database host
--port <PORT> \ # Database port
--user <USER> \ # Database user
--password <PASSWORD> \ # Database password
--dbname <DB_NAME> # Database name
To encourage your agent to reference its archival memory, we recommend adding phrases like "search your archival memory..." for the best results.
You can view loaded data source with:
memgpt list sources
To use MemGPT with Azure, expore the following variables and then re-run memgpt configure
:
# see https://github.com/openai/openai-python#microsoft-azure-endpoints
export AZURE_OPENAI_KEY = ...
export AZURE_OPENAI_ENDPOINT = ...
export AZURE_OPENAI_VERSION = ...
# set the below if you are using deployment ids
export AZURE_OPENAI_DEPLOYMENT = ...
export AZURE_OPENAI_EMBEDDINGS_DEPLOYMENT = ...
Note: your Azure endpoint must support functions or you will get an error. See cpacker#91 for more information.
To use custom endpoints, run export OPENAI_API_BASE=<MY_CUSTOM_URL>
and then re-run memgpt configure
to set the custom endpoint as the default endpoint.
Debugging command not found
If you get command not found
(Linux/MacOS), or a CommandNotFoundException
(Windows), the directory where pip installs scripts is not in your PATH. You can either add that directory to your path (pip show pip | grep Scripts
) or instead just run:
python -m memgpt
Building from source
Clone this repo: git clone https://github.com/cpacker/MemGPT.git
Using poetry:
- Install poetry:
pip install poetry
- Run
poetry install
- Run
poetry run memgpt
Using pip:
- Run
pip install -e .
- Run
python3 main.py
If you're using Azure OpenAI, set these variables instead:
# see https://github.com/openai/openai-python#microsoft-azure-endpoints
export AZURE_OPENAI_KEY = ...
export AZURE_OPENAI_ENDPOINT = ...
export AZURE_OPENAI_VERSION = ...
# set the below if you are using deployment ids
export AZURE_OPENAI_DEPLOYMENT = ...
export AZURE_OPENAI_EMBEDDINGS_DEPLOYMENT = ...
# then use the --use_azure_openai flag
memgpt --use_azure_openai
To create a new starter user or starter persona (that MemGPT gets initialized with), create a new .txt
file in ~/.memgpt/humans
or ~/.memgpt/personas
, then use the --persona
or --human
flag when running main.py
. For example:
# assuming you created a new file ~/.memgpt/humans/me.txt
memgpt
# Select me.txt during configuration process
-- OR --
# assuming you created a new file ~/.memgpt/humans/me.txt
memgpt --human me.txt
You can also specify any of the starter users in /memgpt/humans/examples or any of the starter personas in /memgpt/personas/examples.
You can run MemGPT with GPT-3.5 as the LLM instead of GPT-4:
memgpt
# Select gpt-3.5 during configuration process
-- OR --
memgpt --model gpt-3.5-turbo
Note that this is experimental gpt-3.5-turbo support. It's quite buggy compared to gpt-4, but it should be runnable.
Please report any bugs you encounter regarding MemGPT running on GPT-3.5 to cpacker#59.
You can run MemGPT with local LLMs too. See instructions here and report any bugs/improvements here cpacker#67.
--first
allows you to send the first message in the chat (by default, MemGPT will send the first message)
--debug
enables debugging output
Configure via legacy flags
--model
select which model to use ('gpt-4', 'gpt-3.5-turbo-0613', 'gpt-3.5-turbo')
--persona
load a specific persona file
--human
load a specific human file
--archival_storage_faiss_path=<ARCHIVAL_STORAGE_FAISS_PATH>
load in document database (backed by FAISS index)
--archival_storage_files="<ARCHIVAL_STORAGE_FILES_GLOB_PATTERN>"
pre-load files into archival memory
--archival_storage_files_compute_embeddings="<ARCHIVAL_STORAGE_FILES_GLOB_PATTERN>"
pre-load files into archival memory and also compute embeddings for embedding search
--archival_storage_sqldb=<SQLDB_PATH>
load in SQL database
These are the commands for the CLI, not the Discord bot! The Discord bot has separate commands you can see in Discord by typing /
.
While using MemGPT via the CLI (not Discord!) you can run various commands:
//
toggle multiline input mode
/exit
exit the CLI
/save
save a checkpoint of the current agent/conversation state
/load
load a saved checkpoint
/dump <count>
view the last <count> messages (all if <count> is omitted)
/memory
print the current contents of agent memory
/pop <count>
undo the last messages in the conversation. It defaults to 3, which usually is one turn around in the conversation
/retry
pops the last answer and tries to get another one
/rethink <text>
will replace the inner dialog of the last assistant message with the <text> to help shaping the conversation
/rewrite
will replace the last assistant answer with the given text to correct or force the answer
/heartbeat
send a heartbeat system message to the agent
/memorywarning
send a memory warning system message to the agent
MemGPT's archival memory let's you load your database and talk to it! To motivate this use-case, we have included a toy example.
Consider the test.db
already included in the repository.
id | name | age |
---|---|---|
1 | Alice | 30 |
2 | Bob | 25 |
3 | Charlie | 35 |
To talk to this database, run:
memgpt --archival_storage_sqldb=memgpt/personas/examples/sqldb/test.db
And then you can input the path to your database, and your query.
Please enter the path to the database. test.db
...
Enter your message: How old is Bob?
...
๐ค Bob is 25 years old.
MemGPT enables you to chat with your data locally -- this example gives the workflow for loading documents into MemGPT's archival memory.
To run our example where you can search over the SEC 10-K filings of Uber, Lyft, and Airbnb,
-
Download the .txt files from Hugging Face and place them in
memgpt/personas/examples/preload_archival
. -
In the root
MemGPT
directory, runmemgpt --archival_storage_files="memgpt/personas/examples/preload_archival/*.txt" --persona=memgpt_doc --human=basic
If you would like to load your own local files into MemGPT's archival memory, run the command above but replace --archival_storage_files="memgpt/personas/examples/preload_archival/*.txt"
with your own file glob expression (enclosed in quotes).
In the root MemGPT
directory, run
memgpt main.py --archival_storage_files_compute_embeddings="<GLOB_PATTERN>" --persona=memgpt_doc --human=basic
This will generate embeddings, stick them into a FAISS index, and write the index to a directory, and then output:
To avoid computing embeddings next time, replace --archival_storage_files_compute_embeddings=<GLOB_PATTERN> with
--archival_storage_faiss_path=<DIRECTORY_WITH_EMBEDDINGS> (if your files haven't changed).
If you want to reuse these embeddings, run
memgpt --archival_storage_faiss_path="<DIRECTORY_WITH_EMBEDDINGS>" --persona=memgpt_doc --human=basic
MemGPT also enables you to chat with docs -- try running this example to talk to the LlamaIndex API docs!
-
a. Download LlamaIndex API docs and FAISS index from Hugging Face.
# Make sure you have git-lfs installed (https://git-lfs.com) git lfs install git clone https://huggingface.co/datasets/MemGPT/llamaindex-api-docs mv llamaindex-api-docs
-- OR --
b. Build the index:
- Build
llama_index
API docs withmake text
. Instructions here. Copy over the generated_build/text
folder tomemgpt/personas/docqa
. - Generate embeddings and FAISS index.
cd memgpt/personas/docqa python3 scrape_docs.py python3 generate_embeddings_for_docs.py all_docs.jsonl python3 build_index.py --embedding_files all_docs.embeddings.jsonl --output_index_file all_docs.index
- Build
-
In the root
MemGPT
directory, runmemgpt --archival_storage_faiss_path=<ARCHIVAL_STORAGE_FAISS_PATH> --persona=memgpt_doc --human=basic
where
ARCHIVAL_STORAGE_FAISS_PATH
is the directory whereall_docs.jsonl
andall_docs.index
are located. If you downloaded from Hugging Face, it will bememgpt/personas/docqa/llamaindex-api-docs
. If you built the index yourself, it will bememgpt/personas/docqa
.
If you have any further questions, or have anything to share, we are excited to hear your feedback!
- By default MemGPT will use
gpt-4
, so your API key will requiregpt-4
API access - For issues and feature requests, please open a GitHub issue or message us on our
#support
channel on Discord
Datasets used in our paper can be downloaded at Hugging Face.
- Release MemGPT Discord bot demo (perpetual chatbot)
- Add additional workflows (load SQL/text into MemGPT external context)
- Integration tests
- Integrate with AutoGen (discussion)
- Add official gpt-3.5-turbo support (discussion)
- CLI UI improvements (issue)
- Add support for other LLM backends (issue, discussion)
- Release MemGPT family of open models (eg finetuned Mistral) (discussion)
Reminder: if you do not plan on modifying the source code, simply install MemGPT with pip install pymemgpt
!
First, install Poetry using the official instructions here.
Then, you can install MemGPT from source with:
git clone [email protected]:cpacker/MemGPT.git
poetry shell
poetry install
We recommend installing pre-commit to ensure proper formatting during development:
pip install pre-commit
pre-commit install
pre-commit run --all-files
We welcome pull requests! Please run the formatter before submitting a pull request:
poetry run black . -l 140