A SQLite extension for generating text embeddings from remote APIs (OpenAI, Nomic, Cohere, llamafile, Ollama, etc.). A sister project to sqlite-vec
and sqlite-lembed
. A work-in-progress!
.load ./rembed0
INSERT INTO temp.rembed_clients(name, options)
VALUES ('text-embedding-3-small', 'openai');
select rembed(
'text-embedding-3-small',
'The United States Postal Service is an independent agency...'
);
The temp.rembed_clients
virtual table lets you "register" clients with pure INSERT INTO
statements. The name
field is a unique identifier for a given client, and options
allows you to specify which 3rd party embedding service you want to use.
In this case, openai
is a pre-defined client that will default to OpenAI's https://api.openai.com/v1/embeddings
endpoint and will source your API key from the OPENAI_API_KEY
environment variable. The name of the client, text-embedding-3-small
, will be used as the embeddings model.
Other pre-defined clients include:
Client name | Provider | Endpoint | API Key |
---|---|---|---|
openai |
OpenAI | https://api.openai.com/v1/embeddings |
OPENAI_API_KEY |
nomic |
Nomic | https://api-atlas.nomic.ai/v1/embedding/text |
NOMIC_API_KEY |
cohere |
Cohere | https://api.cohere.com/v1/embed |
CO_API_KEY |
jina |
Jina | https://api.jina.ai/v1/embeddings |
JINA_API_KEY |
mixedbread |
MixedBread | https://api.mixedbread.ai/v1/embeddings/ |
MIXEDBREAD_API_KEY |
llamafile |
llamafile | http://localhost:8080/embedding |
None |
ollama |
Ollama | http://localhost:11434/api/embeddings |
None |
Different client options can be specified with remebed_client_options()
. For example, if you have a different OpenAI-compatible service you want to use, then you can use:
INSERT INTO temp.rembed_clients(name, options) VALUES
(
'xyz-small-1',
rembed_client_options(
'format', 'openai',
'url', 'https://api.xyz.com/v1/embeddings',
'key', 'xyz-ca865ece65-hunter2'
)
);
Or to use a llamafile server that's on a different port:
INSERT INTO temp.rembed_clients(name, options) VALUES
(
'xyz-small-1',
rembed_client_options(
'format', 'lamafile',
'url', 'http://localhost:9999/embedding'
)
);
sqlite-rembed
works well with sqlite-vec
, a SQLite extension for vector search. Embeddings generated with rembed()
use the same BLOB format for vectors that sqlite-vec
uses.
Here's a sample "semantic search" application, made from a sample dataset of news article headlines.
create table articles(
headline text
);
-- Random NPR headlines from 2024-06-04
insert into articles VALUES
('Shohei Ohtani''s ex-interpreter pleads guilty to charges related to gambling and theft'),
('The jury has been selected in Hunter Biden''s gun trial'),
('Larry Allen, a Super Bowl champion and famed Dallas Cowboy, has died at age 52'),
('After saying Charlotte, a lone stingray, was pregnant, aquarium now says she''s sick'),
('An Epoch Times executive is facing money laundering charge');
-- Build a vector table with embeddings of article headlines, using OpenAI's API
create virtual table vec_articles using vec0(
headline_embeddings float[1536]
);
insert into vec_articles(rowid, headline_embeddings)
select rowid, rembed('text-embedding-3-small', headline)
from articles;
Now we have a regular articles
table that stores text headlines, and a vec_articles
virtual table that stores embeddings of the article headlines, using OpenAI's text-embedding-3-small
model.
To perform a "semantic search" on the embeddings, we can query the vec_articles
table with an embedding of our query, and join the results back to our articles
table to retrieve the original headlines.
param set :query 'firearm courtroom'
with matches as (
select
rowid,
distance
from vec_articles
where headline_embeddings match rembed('text-embedding-3-small', :query)
order by distance
limit 3
)
select
headline,
distance
from matches
left join articles on articles.rowid = matches.rowid;
/*
+--------------------------------------------------------------+------------------+
| headline | distance |
+--------------------------------------------------------------+------------------+
| The jury has been selected in Hunter Biden's gun trial | 1.05906391143799 |
+--------------------------------------------------------------+------------------+
| Shohei Ohtani's ex-interpreter pleads guilty to charges rela | 1.2574303150177 |
| ted to gambling and theft | |
+--------------------------------------------------------------+------------------+
| An Epoch Times executive is facing money laundering charge | 1.27144026756287 |
+--------------------------------------------------------------+------------------+
*/
Notice how "firearm courtroom" doesn't appear in any of these headlines, but it can still figure out that "Hunter Biden's gun trial" is related, and the other two justice-related articles appear on top.
- No batch support yet. If you use
rembed()
in a batch UPDATE or INSERT in 1,000 rows, then 1,000 HTTP requests will be made. Add a 👍 to Issue #1 if you want to see this fixed. - No builtin rate limiting. Requests are sent sequentially so this may not come up in small demos, but
sqlite-rembed
could add features that handles rate limiting/retries implicitly. Add a 👍 to Issue #2 if you want to see this implemented.