Helicone on Railway

Open-source LLM observability that logs cost, latency and prompts, self-hosted on Railway.

Official site · License: Apache-2.0 · 2 Railway templates

Deploy Helicone on RailwayCompare templates

Helicone is an open-source LLM observability platform. It logs requests to model providers with cost, latency, token counts and the full prompts and responses, and groups them into sessions so you can debug agents and track spend. Apps send traffic through a Helicone gateway, or log to it, using a Helicone API key.

It is used by teams that want to see what their application sends to OpenAI or Anthropic and what it costs, without keeping that data with a third party when self-hosted.

Helicone is released under the Apache-2.0 licence.

What Helicone can do

  • Request logging with cost, latency and token usage
  • Full prompt and response bodies
  • Sessions for tracing agent runs
  • Gateway for OpenAI and Anthropic calls
  • Dashboard with API keys and settings
  • Analytics stored in ClickHouse

What people use it for

  • Logging LLM calls with cost, latency and token usage
  • Debugging agents with sessions and full request bodies
  • Tracking spend across OpenAI and Anthropic
  • Keeping prompts and completions inside your own project

Videos

Helicone AI — The Open-source LLM Observability for Developers | Product Hunt · Helicone AI
How to monitor LLM requests from Open WebUI · Helicone AI

Running Helicone on Railway

The Railway template deploys the self-hosted stack at release v2025.08.21 behind a Caddy router on one public domain. It runs the Helicone dashboard and the Jawn API server, a migrations service, ClickHouse for analytics, Postgres for accounts and API keys, and SeaweedFS for request and response bodies. ClickHouse and Postgres use volumes. An admin account is created automatically and public sign-up is closed. Railway bills the resources the services use.

Helicone templates on Railway

Helicone LLM Observability GatewayMaintained by Tracklayer
Helicone is an open-source LLM observability platform that logs requests with cost, latency, tokens and full prompts and responses. This template deploys the self-hosted stack (rel
2 deployments
Deploy
Helicone by an independent creator
Open-source LLM observability: log cost, latency and prompts of AI calls
0 deployments
Deploy

Tracklayer templates are listed first. Other Helicone templates follow, ordered by total deployments on Railway. How this works.

Deployment numbers as of 2026-11-20.

Questions about Helicone

How do I log in?

Enter your email in HELICONE_ADMIN_EMAIL at deploy. When the migrations service logs that it finished, open /signin on the proxy domain and use that email with HELICONE_ADMIN_PASSWORD from the migrations service variables.

How do I send traffic through it?

Create a Helicone API key in Settings. Point OpenAI clients at https://{proxy-domain}/jawn/v1/gateway/oai/v1, or Anthropic clients at /jawn/v1/gateway/anthropic, with a Helicone-Auth Bearer header. Your provider key stays in your app.

How do I add teammates?

Public sign-up is closed. Set HELICONE_ALLOW_SIGNUP to true on the proxy service while teammates register, then switch it back. E-mails need SMTP, which requires a Railway Pro plan.

How do I upgrade?

Upgrade the web, jawn and migrations images together to a newer release tag and redeploy. The migrations job applies schema changes. ClickHouse is pinned in services/clickhouse/Dockerfile.

What are the scaling limits?

jawn and web are stateless and can run more replicas. ClickHouse and Postgres scale vertically.

Guides

Similar apps

Tracklayer maintains the templates marked as such; Helicone is a trademark of its owners and this site is not affiliated with them.