# TEI - Text Embeddings Inference

Lightweight embedding service for Clawdbot memory search.

## Overview

Runs [HuggingFace Text Embeddings Inference](https://github.com/huggingface/text-embeddings-inference) (TEI) on CPU to provide OpenAI-compatible embeddings for semantic memory search.

- **Model:** `BAAI/bge-base-en-v1.5` (~440MB, 768 dimensions, 512 token context)
- **Endpoint:** `http://text-embeddings:8080/v1/embeddings`
- **No GPU required**

## Deploy

```bash
oc new-project tei
oc apply -f deployment.yaml
```

## Clawdbot Config

Add to `~/.clawdbot/clawdbot.json`:

```json5
{
  "agents": {
    "defaults": {
      "memorySearch": {
        "enabled": true,
        "provider": "openai",
        "model": "BAAI/bge-base-en-v1.5",
        "remote": {
          "baseUrl": "http://text-embeddings.tei.svc.cluster.local:8080/v1/",
          "apiKey": "not-needed"
        }
      }
    }
  }
}
```

## Resources

- **Memory:** 512Mi request, 1Gi limit
- **CPU:** 250m request, 1000m limit
- **Storage:** 1Gi PVC for model cache

## Test

```bash
curl -X POST http://text-embeddings:8080/v1/embeddings \
  -H "Content-Type: application/json" \
  -d '{"input": "Hello world", "model": "BAAI/bge-base-en-v1.5"}'
```