Free during beta. Self-hosted is free under Apache 2.0 license. Enterprise options available on request.
Pay-per-unit pricing with prepaid credits. Image models start at 0.4 credits per image (FLUX.1 schnell) to 5 credits per image (GPT Image 1). Video models range from 7.5 credits per second (Seedance 1.0 Pro) to 48 credits per second (Veo 3 Fast). Audio models are 12 credits per 1K characters for voi
Key features
Prompt versioning with named aliases and diffs
Server-side prompt templating with conditionals and loops
OpenAI-compatible AI gateway with automatic model fallbacks
Virtual keys, budgets and per-key rate limits
LLM tracing through the gateway or over OpenTelemetry
Versioned tool catalog for function calling
Offline experiments and online evaluation rules
Prompt optimization with candidate rewrites
Team-wide audit trail of every change
MCP server for Claude Code, Claude Desktop, Cursor
TypeScript and Python SDKs with built-in retries, idempotency keys, streaming and typed errors
Support for 15 models from leading providers (OpenAI, Anthropic, Google, Meta, etc.)
Unified API with one key and one bill for all models
Customer tracking and per-customer usage reporting
Monthly spend limits and low-balance alerts
Automatic failover to alternative providers when available
OpenAI-compatible chat API endpoint
What makes it different
Change prompts without a deploy by moving production alias