> ## Documentation Index
> Fetch the complete documentation index at: https://docs.buildpersona.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# LLM Providers

> Configuration for different providers

Persona supports multiple LLM providers. The format is `provider/model`.

## Supported Providers

| Provider         | Format               | Example                   |
| ---------------- | -------------------- | ------------------------- |
| OpenAI           | `openai/{model}`     | `openai/gpt-5.2`          |
| Azure OpenAI     | `azure/{deployment}` | `azure/gpt-4o`            |
| Azure AI Foundry | `foundry/{model}`    | `foundry/gpt-5.2`         |
| Anthropic        | `anthropic/{model}`  | `anthropic/claude-3-opus` |
| Gemini           | `gemini/{model}`     | `gemini/gemini-pro`       |

## Configuration

Set your provider in `.env`:

```env theme={null}
LLM_SERVICE=openai/gpt-5.2
EMBEDDING_SERVICE=openai/text-embedding-3-small
```

See `.env.example` for the full reference.

## OpenAI (Recommended)

The easiest setup for new users:

```env theme={null}
LLM_SERVICE=openai/gpt-5.2
EMBEDDING_SERVICE=openai/text-embedding-3-small
OPENAI_API_KEY=sk-your-key-here
```

## Azure AI Foundry

For enterprise/production deployments:

```env theme={null}
LLM_SERVICE=foundry/gpt-5.2
AZURE_API_KEY=your-azure-key
AZURE_API_BASE=https://your-resource.openai.azure.com
AZURE_CHAT_DEPLOYMENT=gpt-5.2
AZURE_EMBEDDING_DEPLOYMENT=text-embedding-3-small
```

## Neo4j Database

The database URI depends on your environment:

```env theme={null}
# Running inside Docker (docker compose up):
URI_NEO4J=bolt://neo4j:7687

# Running locally (poetry run python ...):
URI_NEO4J=bolt://localhost:7687

USER_NEO4J=neo4j
PASSWORD_NEO4J=password
NEO4J_AUTH=neo4j/password
```

## Performance Tuning

```env theme={null}
# Parallel session ingestion (for multi-session batch ingestion)
INGEST_SESSION_CONCURRENCY=5

# Eval judge model (for evals only)
EVAL_JUDGE_MODEL=gpt-5-mini
```

## Embedding Fallback

If your primary provider doesn't support embeddings, Persona falls back to OpenAI's embedding API. Ensure `OPENAI_API_KEY` is set even when using other providers.