Using more than one model in processing

[LStromann]

=I'm using environment variables to define the model, but can I change it during processing? I'd like to use more than one model in the Cognee pipeline, for example, using a model-tuned for a specific purpose and then using a generic model after that.

This discussion was automatically pulled from Discord.

[LStromann]

Using more than one model in processing

[LStromann]

Not currently, but for now you can run multiple Cognee instances with different models (eg. one Cognee instance for search and another one for cognification )

And call the appropriate instance based on the need

Provided you are using server based databases (for example Neo4j, PostgreSQL and PGVector) and not file based like (Kuzu, LanceDB and SQLite)

[LStromann]

or you can setup a proxy server for LLM requests that will sort which model to call but that is a bit harder and would require more time to setup

[LStromann]

Thank you <@508682252595626015>. Can I "pass" the result of .add() to .cognify() or .memify() from different instances? Does each instance need to use the same databases?

[LStromann]

Each instance needs to share the same databases

[LStromann]

and then the results will be available on both instances

[LStromann]

ok, I got it. Thank you.

[LStromann]

<@508682252595626015> curious, why can't we change the model? You should be able to overwrite a model in a custom pipeline by setting llm_api_key variable as you go and overriding config default one

[LStromann]

Right now our LLM config is global so if it's overriden somewhere it will be overriden for the whole instance in pipelines/search etc., we can make it work like with the database configurations so it's a context config but it would take some time to do and refactor