AI Providers

LenserFight now treats providers as catalog entries with explicit support tiers instead of assuming every documented provider is directly runnable.

Support tiers

Tier	Meaning
runnable	LenserFight has a direct runtime path or adapter for this provider.
byok_only	Users can configure credentials, but execution depends on BYOK routing rather than a fully native runtime path.
catalog_only	The provider appears in discovery, planning, and showroom UIs but is not executed directly.
deprecated	The provider remains listed for compatibility or migration context and should not be chosen for new workflows.

Runtime-first providers

These providers are part of the primary execution surface today:

• openai
• anthropic
• google
• mistral
• ollama
• fal
• stability
• elevenlabs
• kling
• suno

Catalog and gateway providers

These providers are represented in the catalog for comparison, workflow planning, gateway routing, or future adapter work:

• openrouter
• perplexity
• xai
• groq
• deepseek
• bedrock
• runway
• litellm
• lmstudio

Deprecated providers

• midjourney — no public API available; use fal or stability for image generation

Provider metadata carried by the catalog

Each provider entry is expected to carry:

• canonical key
• display name
• docs URL
• support tier
• logo slug for cards and sidebars
• gateway/platform compatibility metadata
• source provenance metadata such as source_url and source_checked_at

Product usage

The provider catalog feeds four surfaces:

1. Supabase seed data in ai.providers
2. The human-facing AI catalog showroom at /ai/catalog
3. Agent control-room provider and model pages under /lenser/:handle/ag/*
4. CLI commands such as providers list, providers show, providers config, and gateway