Alpha Engine & Models.

The platform runs on a 3-provider AI stack with automatic failover: Anthropic (primary) → Google Gemini (fallback 1) → OpenAI (fallback 2). This guarantees 99.99% AI uptime.

Primary — Claude (Anthropic)

All agents run on Claude by default. Claude powers the chat endpoint, the orchestrator's intent classification, the Alpha Engine's reasoning, and the memory extraction pipeline. Models: claude-3-7-sonnet-latest (smart), claude-3-7-sonnet-latest (balanced), claude-3-7-sonnet-latest (fast). 200K context window.

Fallback 1 — Google Gemini

If Anthropic returns a 5xx error or rate limit, the circuit breaker automatically switches to Gemini. Models: gemini-1.5-pro, gemini-1.5-flash, gemini-2.0-flash. 1M context window. Full streaming and tool calling support. Requires GOOGLE_API_KEY.

Fallback 2 — OpenAI

If both Anthropic and Gemini are unavailable, OpenAI is the final fallback. Models: gpt-4o (smart), gpt-4o-mini (fast). 128K context window. Requires OPENAI_API_KEY.

Circuit breaker

Each provider has an independent circuit breaker with a 100-call sliding window. The breaker opens at >30% failure rate and stays open for 60 seconds before retrying. Failover is transparent — users never see provider errors.

BYOK — Bring Your Own Key

Add your own Anthropic API key in Settings → API Keys. It is stored in users.preferences.anthropicApiKey. The chat route resolves keys in priority order: platform env key → user BYOK key → demo mode. BYOK gives you full control over model selection and costs.

Alpha Engine models

The Alpha Engine at /api/alpha uses specialized system prompts for high-stakes reasoning: sales negotiation, legal analysis, competitive strategy. These calls use claude-3-7-sonnet-latest for quality (scripts, contracts, strategy) and claude-3-7-sonnet-latest for speed (lead scoring).