Vector Dimension Mismatch Error When Switching to gpt-4o-mini in SQLite Database

[rferrari]

Describe the bug

Using an SQLite database, after changing models to gpt-4o-mini, I encountered the following vector error:

I started using the default OpenAI models, then switched the MEDIUM and LARGE models to gpt-4o-mini and encountered this error.

---

To Reproduce

1. Use an SQLite database as the storage backend.
2. Start with the default OpenAI models.
3. Change the MEDIUM and LARGE models to gpt-4o-mini.
4. Attempt to start agents.

---

Expected behavior

The system should seamlessly support model changes without vector dimension mismatches or database errors.

---

Screenshots

["⛔ SqliteError: Vector dimension mismatch. First vector has 384 dimensions, while the second has 1536 dimensions."]

⛔ ERRORS
Error starting agents:
{"code":"SQLITE_ERROR"}

---

Additional context

It seems the SQLite database stores vectors with a specific dimension (384 in this case) that is incompatible with the gpt-4o-mini model (1536 dimensions). This may require re-indexing or updating the schema for the new vector size.

[veebull]

get same error when switch llm from openai to another
just removed db.sqlite file in

📁 eliza/
└── 📁 agent/
           └── 📁 data/
                     └── 📄 db.sqlite

and then

pnpm start

and in my case everything works fine

[AIFlowML]

@rferrari the vector shape is wrong in the old DB and is correct 1536 in the request.
When you change models from local to OAI or else the embedding become 1536 standard OAI.
You have to remove the old DB that is not compliant anymore.