Libem is an open-source compound AI toolchain for fast and accurate entity matching, powered by LLMs.

make install

Update ~/.libem/config.yaml with

OPENAI_API_KEY: your key

Optionally, to support browsing:

GOOGLE_CSE_ID: your key

And:

GOOGLE_API_KEY: your key

To run the sample example.

make run

You can also install libem library with pip:

pip install libem

import libem

libem.match("apple", "orange")

In shell:

libem apple orange

Libem enforces a strict naming convention for a toolchain and the tool access methods. Toolchain's main methods are defined in the interface.py and exposed at the toolchain level. For example, the following code will import the tune tool in libem toolchain.

from libem import tune

To import the libem.tune toolchain, we must explicitly invoke:

import libem

tune = libem.toolchain("libem.tune") 

Note that this does not prevent us from importing the tools in the toolchain directly:

from libem.tune import learn

In a nutshell, one can directly import a toolchain's tools but not the tool's module. This is to encourage (but not enforce) that tools are always accessed via well-defined methods consistent with the LM access.