# Multi-Provider LLM Streaming over a Terminal Relay

> How the SessionCast agent streams completions from Anthropic, Claude Code and OpenAI-compatible providers chunk by chunk over WebSocket.

Source: https://blog.sessioncast.io/posts/multi-provider-llm-streaming/ · 2026-02-28 · devload

The CLI agent now supports multiple LLM providers with real-time streaming. Connect your Anthropic API key, Claude Code, or any OpenAI-compatible endpoint, and stream completions through SessionCast.

## Supported Providers

- **anthropic** — Direct Anthropic API (Claude 3.5 Sonnet, Opus, etc.)
- **claude-code** — Claude Code CLI (uses your local Claude Code installation)
- **openai** — OpenAI API (GPT-4, etc.)

## Streaming

LLM responses now stream chunk by chunk over WebSocket. The web viewer and SDK clients receive partial responses as they're generated — no more waiting for the full completion.

```
Client → relay → agent → LLM provider
                           ↓ (streaming)
Client ← relay ← agent ← chunk 1
Client ← relay ← agent ← chunk 2
Client ← relay ← agent ← chunk 3
Client ← relay ← agent ← [done]
```

## Using via Go SDK

```go
stream, err := client.LlmChatStream(ctx, &sessioncast.LlmChatRequest{
    Model: "anthropic",
    Messages: []sessioncast.ChatMessage{
        {Role: "user", Content: "Explain WebSockets in 3 sentences."},
    },
})

for event := range stream {
    if event.Chunk != nil {
        fmt.Print(event.Chunk.Content)
    }
}
```

## Using via Web Viewer

The web viewer's command bar supports LLM chat. Type a question and get a streaming response directly in the terminal interface.

## Provider Configuration

Providers are auto-detected based on environment variables:

- `ANTHROPIC_API_KEY` — enables the `anthropic` provider
- Claude Code installed at `~/.claude` — enables the `claude-code` provider
- `OPENAI_API_KEY` — enables the `openai` provider

The agent negotiates capabilities with connecting clients, so they only see providers that are actually available.

## Requirements

- **CLI Agent** `v2.2.2` or later
- At least one LLM provider configured on the agent machine