Helicone is an open-source AI gateway and LLM observability platform aimed at engineers running LLM-backed apps and agents in production. Integration is a base-URL swap: point an OpenAI-compatible client at the gateway and requests are logged while gaining access to 100+ models across many providers through one API, with intelligent routing and automatic fallbacks. Beyond OpenAI-style clients it integrates with providers like Anthropic, Gemini, Bedrock, and Groq, and with frameworks such as LangChain, LlamaIndex, LangGraph, the Vercel AI SDK, and CrewAI.
On the observability side it captures traces and sessions for agents, chatbots, and document pipelines; tracks cost, latency, and quality metrics (with PostHog export for custom dashboards); and offers a playground for iterating on prompts against real traffic. Prompt management lets you version prompts from production data and deploy them through the gateway without code changes, and datasets can feed fine-tuning via partner services.
It can be used as a hosted service or self-hosted via Docker Compose (with a Helm chart for enterprise deployments). The self-hosted stack comprises a Next.js frontend, a Cloudflare Workers proxy logger, an Express log-collection server, Supabase for the application database and auth, ClickHouse for analytics, and Minio for log storage. The project also maintains an open pricing database covering 300+ models for cost calculation.
