Helicone
Self-hostable LLM proxy that logs cost, latency, and traces, in maintenance mode since a 2026 acquisition.
About
Helicone sits between an application and its model providers as a logging proxy and gateway: point an OpenAI-compatible client at its base URL and each request is recorded with cost, latency, token counts, and user or session tags, and can be routed with fallbacks across OpenAI, Anthropic, Gemini, Bedrock, Groq, Together, and others. The dashboard adds trace and session views for agents, a prompt playground, versioned prompts, datasets, rate limits, alerts, and PostHog export. The whole stack is Apache-2.0 and runs from a docker-compose script: a Next.js web app, a Cloudflare Workers proxy, an Express log server called Jawn, Supabase, ClickHouse, and MinIO; a Helm chart is offered to enterprise customers on request. Important context: Mintlify acquired Helicone on March 3, 2026 and put it in maintenance mode, so security patches, new model support, and bug fixes continue while new features should not be expected. It suits teams already on Helicone and anyone who wants a self-hosted LLM request log under their own control.
Should you use Helicone?
Pick it when
Pick Helicone when you want request logging, response caching, rate limits, and cost tracking by pointing your client at a proxy instead of adding tracing code, with a simple Docker or Helm self-host if needed.
Look elsewhere when
Skip it if you do not want a proxy in the request path: caching and rate limits depend on it, and without them Langfuse or OpenLLMetry instrument in-process just as well. OpenLIT adds GPU monitoring and guardrails.
Alternatives to Helicone
- Portkey AI Gateway
Gateway-first, with automatic retries, fallbacks, and load balancing across providers; pick it when routing matters more than the logging UI.
- Langfuse
SDK-based tracing of nested chains plus evals, with no proxy in the request path; more setup than a base URL change and no response caching.
- LiteLLM
MIT gateway usable as a Python SDK or a proxy, with load balancing, fallbacks, and team spend tracking across 100+ providers; pick it when routing is the need.
- OpenLIT
Self-hosted OTel platform that adds GPU monitoring, guardrails, and evals; instruments via SDK rather than a proxy, so no response caching, and takes more setup.
Reviews (0)
Leave a Review
No reviews yet. Be the first to review!
Details
- Category
- AI Observability & Evaluation
- Price
- Free
- Platform
- Local/Desktop
- Difficulty
- Beginner (1/5)
- License
- Apache-2.0
- Added
- Apr 3, 2026