tsellm 0.1.0a10

Interactive SQLite shell with LLM support

tsellm: Use LLMs in SQLite

tsellm is the easiest way to access LLMs through your SQLite database.

pip install tsellm

Behind the scenes, tsellm is based on the beautiful llm library, so you can use any of its plugins:

Generative

For example, to access gpt4all models

llm install llm-gpt4all
# Then pick any gpt4all (it will be downloaded automatically the first time you use any model
tsellm :memory: "select prompt('What is the capital of Greece?', 'orca-mini-3b-gguf2-q4_0')"
tsellm :memory: "select prompt('What is the capital of Greece?', 'orca-2-7b')"

Embeddings

llm install llm-sentence-transformers
llm sentence-transformers register all-MiniLM-L12-v2
tsellm :memory: "select embed('Hello', 'sentence-transformers/all-MiniLM-L12-v2')"

Embeddings for binary (BLOB) columns

wget https://tselai.com/img/flo.jpg
sqlite3 images.db <<EOF
CREATE TABLE images(name TEXT, type TEXT, img BLOB);
INSERT INTO images(name,type,img) VALUES('flo','jpg',readfile('flo.jpg'));
EOF

llm install llm-clip
tsellm images.db "select embed(img, 'clip') from images"

Examples

Things get more interesting if you combine models in your standard SQLite queries.

First, create a db with some data

sqlite3 prompts.db <<EOF
CREATE TABLE [prompts] (
   [p] TEXT
);
INSERT INTO prompts VALUES('hello world!');
INSERT INTO prompts VALUES('how are you?');
INSERT INTO prompts VALUES('is this real life?');
INSERT INTO prompts VALUES('1+1=?');
EOF

With a single query you can access get prompt responses from different LLMs:

tsellm prompts.db "
        select p,
        prompt(p, 'orca-2-7b'),
        prompt(p, 'orca-mini-3b-gguf2-q4_0'),
        embed(p, 'sentence-transformers/all-MiniLM-L12-v2') 
        from prompts"

Interactive Shell

If you don't provide an SQL query, you'll enter an interactive shell instead.

tsellm prompts.db

Installation

pip install tsellm

How

tsellm relies on the following facts:

• SQLite is bundled with the standard Python library (import sqlite3)
• Python 3.12 ships with a SQLite interactive shell
• one can create Python-written user-defined functions to be used in SQLite queries (see create_function)
• Simon Willison has gone through the process of creating the beautiful llm Python library and CLI

Development

pip install -e '.[test]'
pytest