Helicone is an open-source LLM observability platform and AI gateway for AI engineers, backed by YC W23. Its core purpose is to help developers log, monitor, debug, and experiment with AI applications, while simplifying multi-model access through a unified gateway.
Main capabilities:
- AI Gateway: Access 100+ AI models through a single API key, compatible with the OpenAI API, supporting intelligent routing and automatic fallback.
- Quick integration: Log requests from OpenAI, Anthropic, LangChain, Gemini, Vercel AI SDK, etc. with one line of code.
- Observability: Inspect and debug traces and sessions for agents, chatbots, document processing pipelines, etc.
- Analytics: Track metrics such as cost, latency, and quality, and export to PostHog with one line of code to build custom dashboards.
- Playground: Quickly test and iterate on prompts, sessions, and traces in the UI.
- Prompt management: Version prompts using production data, and deploy prompts through the AI gateway without changing code; prompts always remain under the user's control and accessible at any time.
- Fine-tuning: Fine-tune with fine-tuning partners such as OpenPipe or Autonomi (more coming soon).
- Enterprise-ready: SOC 2 and GDPR compliant.
- Free tier: A free tier of 10k requests per month, no credit card required.
How to use: Sign up to get an API key and add credits at helicone.ai/credits. Then update the baseURL in your code and add the API key. Example: import OpenAI from "openai"; const client = new OpenAI({ baseURL: "https://ai-gateway.helicone.ai", apiKey: process.env.HELICONE_API_KEY }); const response = await client.chat.completions.create({ model: "gpt-4o-mini", messages: [{ role: "user", content: "Hello!" }] }); Supported models include claude-sonnet-4, gemini-2.0-flash, or any model from https://www.helicone.ai/models . After setup, you can view logs in Helicone.
Self-hosting: Helicone is easy to self-host and update. You can start it locally using the docker-compose file. For enterprise workloads, you can use the production-grade Helm chart, which requires contacting enterprise@helicone.ai to obtain. Manual deployment is not recommended; please use Docker or Helm. Helicone consists of five services: Web (frontend platform, NextJS), Worker (proxy logging, Cloudflare Workers), Jawn (dedicated server for collecting logs, Express + Tsoa), Supabase (application database and authentication), ClickHouse (analytics database), Minio (log object storage).
Inference providers: The integration list may be outdated; check the documentation for the latest integrations. If you cannot find one, you can contact help@helicone.ai to request a new integration.
Use cases: Suitable for AI engineers who need to monitor and debug AI applications, and for teams that want to access multiple models through a unified gateway, manage prompt versions, and analyze metrics such as cost and latency.
Limitations: The material does not provide specific pricing, paid plan prices, or billing details beyond the free tier. Manual deployment is not recommended. The integration list may be outdated and the documentation should be considered authoritative.
