Use AnythingLLM with amazee.ai
AnythingLLM is a full-stack, self-hosted AI application for document chat, RAG, and AI agents. Connect it to amazee.ai to use Claude and other models with your documents, all processed in your chosen region.
Prerequisites
- AnythingLLM installed (installation guide)
- An amazee.ai API key and endpoint from my.amazee.io
Connect amazee.ai as the LLM provider
- Open AnythingLLM and go to Settings → LLM Preference
- Select Generic OpenAI (or Custom OpenAI API compatible depending on your version)
- Fill in:
| Field | Value |
|---|---|
| Base URL | https://llm.[your-region].amazee.ai/v1 |
| API Key | your amazee.ai API key |
| Model | claude-sonnet-4-5 (or your preferred model) |
| Token context window | 200000 for Claude Sonnet/Opus; 200000 for Haiku |
- Click Save changes
See Available Models for the full list.
Connect pgvector for embeddings (optional)
AnythingLLM can use pgvector as its vector database. To get pgvector credentials for your workspace, contact ai.support@amazee.io.
- Go to Settings → Vector Database
- Select PGVector
- Fill in:
| Field | Value |
|---|---|
| Connection string | postgresql://USERNAME:PASSWORD@vectordb.[your-region].amazee.ai:5432/DATABASE |
- Click Save changes
Embedding model
Go to Settings → Embedding Preference, select Generic OpenAI, and use:
| Field | Value |
|---|---|
| Base URL | https://llm.[your-region].amazee.ai/v1 |
| API Key | your amazee.ai API key |
| Model | amazon.titan-embed-text-v2:0 |
Verify
Upload a document to a workspace and ask a question about it. A working setup returns an answer grounded in the document's content.
Troubleshooting
"LLM did not respond" errors
: Check that the base URL includes /v1 and the model ID exactly matches what's in Available Models.
Embeddings not working : Make sure the embedding model field is set — AnythingLLM won't use the chat model for embeddings automatically.
Slow responses on large documents : This is normal for large context windows. Claude models support up to 200k tokens. For very large documents, consider chunking them into multiple uploads.