Helicone
AI gateway and LLM observability platform that logs requests, cost and latency and routes to 100+ models.
One base URL change to log, route and cache LLM calls, with Docker self-hosting.
Helicone is an AI gateway and LLM observability platform: you point your OpenAI-style client at its gateway and every request is logged with cost and latency. It suits developers who want proxy-style logging and model routing more than deep evaluation tooling, and who accept a heavy self-hosting stack.
What it is
Helicone sits between your app and providers such as OpenAI, Anthropic, Azure OpenAI, Bedrock and Gemini, giving access to 100+ models through one API key with fallbacks. It records requests, sessions and traces, tracks cost and latency, and adds a playground and prompt versioning that can be deployed through the gateway without code changes.
Who it's for
It suits small teams and agent builders who want to switch one baseURL line and immediately see spend and latency per request. It is a weaker fit if you need dataset-driven evaluation as the centre of your workflow, and the repository's own release page shows no new release since August 2025.
Key features
One-line gateway integration
Change the baseURL and API key in your OpenAI-style client to route calls through the AI Gateway.
Routing with fallbacks
Reach 100+ models with one key and automatic fallbacks between them.
Sessions and agent traces
Inspect and debug sessions for agents, chatbots and document-processing pipelines.
Cost and latency tracking
Track spend, latency and quality metrics, and export to PostHog in one line.
Gateway prompt deployment
Version prompts from production data and deploy them through the gateway without code changes.
Fine-tuning partner hand-off
Send logged data on to fine-tuning partners OpenPipe or Autonomi.
Up and running in 3 steps
Our honest verdict
What's great
- Switching to the gateway is a base URL and key change, shown in the README as a few lines of TypeScript.
- The gateway combines logging with routing to many models and automatic fallbacks, which pure tracing tools do not do.
- Cost tracking is a headline feature, with a documented method for how cost is calculated.
- The cloud free tier is 10k requests a month with no credit card, so you can try it before self-hosting.
Where it falls short
- Self-hosting means five services: a NextJS web app, a Cloudflare Workers proxy, the Jawn log server, Supabase for the app database and auth, ClickHouse for analytics, plus Minio for object storage.
- The production Helm chart is not in the open: the README says to email Helicone's enterprise address to get it, and manual deployment is marked not recommended.
- The latest release listed on the GitHub releases page is dated August 2025, and the pricing page now says Helicone is part of Experiential Labs with details to come, so check the project's direction before committing.
- Evaluation is lighter than in eval-first tools; the README lists prompts, playground and datasets, with RAGAS as an external integration.
Pricing & license
The code is free under Apache-2.0; self-hosting costs you the servers. Helicone's cloud pricing (helicone.ai/pricing) shows Hobby free with 10,000 requests, 1 GB storage and 7-day retention; Pro at $79 a month and Team at $799 a month, both with usage-based pricing on top and a 7-day free trial; and Enterprise on custom terms with SAML SSO and on-prem deployment.
Open remote AI-training roles
767 open remote AI-training roles, updated November 14, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Can I run Helicone without the cloud?
Yes, the README gives a Docker Compose route. It starts multiple services including ClickHouse, Supabase and Minio, so expect to run a small stack.
Is Helicone a proxy or an SDK?
Its main route is the AI Gateway proxy where you change the base URL, and it also lists async logging through OpenLLMetry.
Is Helicone still actively developed?
The GitHub releases page shows its last release in August 2025, and the pricing page says Helicone is now part of Experiential Labs with details to come.
