💡Support Local Embeddings

[MarkSchmidty]

Duplicates

• I have searched the existing issues

Summary 💡

Local Models (LLaMA & its finetunes) now work in a fork of Auto-GPT, including with Pinecone Embeddings. See #25 (comment)

Local models and embeddings offer better privacy, lower costs, and enable new uses, like Auto-GPT experiments in private/air-gapped networks. To get these benefits, we should add local (offline) embeddings storage and recall to Auto-GPT.

Examples 🌈

A version of ooba's text-generation-webui, called wawawario2/long_term_memory, has done this using zarr and Numpy. Check wawawario2/long_term_memory#how-it-works-behind-the-scenes

Though the Auto-GPT fork uses ooba's webui API for local models, the long_term_memory project is closely tied to ooba's UI. We mention it only as a reference. We need to create a similar setup in Auto-GPT.

Motivation 🔦

1. Better Privacy: Local embeddings keep user data on their devices or networks, avoiding the risk of online data breaches.
2. Lower Costs: Users save money on cloud storage, bandwidth, and processing power when they use local embeddings.
3. Offline Use: Auto-GPT can work in offline environments or areas with limited internet, like secure facilities or remote research stations.
4. Customization: Users can create and manage their own embeddings for better model performance and results.
5. Faster Response: Local embeddings can speed up response times since data doesn't need to travel to remote servers.

By adding local embeddings storage and recall to Auto-GPT, users get more control, flexibility, and benefits like privacy, cost savings, and accessibility.

[MarkSchmidty]

This feature is blocking Fully Air-Gapped Offline Auto-GPT

[slavakurilyak]

Related issue #273

[9cento]

+1

[github-actions]

This issue was closed automatically because it has been stale for 10 days with no activity.