Helicone is an open-source AI gateway that logs, traces, and optimizes every LLM API call with one line of code.
Helicone is an open-source AI gateway and LLM observability platform that lets developers monitor, debug, and control every call their application makes to a language model. Instead of adding tracing code throughout a codebase, teams point their existing OpenAI, Anthropic, or other API requests at Helicone’s proxy by changing one line, the base URL.
Every request that passes through gets logged automatically, along with its cost, latency, and token count. For AI developers shipping production systems, that visibility matters. Without it, a single runaway prompt or an unexpected model upgrade can quietly blow through a budget or degrade response quality, and nobody notices until a customer complains.
Helicone sits between your application and the LLM provider as a proxy, a lightweight server that intercepts requests before they reach their destination. You change the base URL in your API client to point at Helicone instead of OpenAI or Anthropic directly, and Helicone forwards the request, logs it, and returns the response unchanged.
Here is what happens on each call:
Request logging: Helicone captures the full request and response, including prompts, completions, and any custom metadata you attach.
Cost and token tracking: It calculates token counts and dollar cost per call across more than 100 supported models, so you see spend by user, feature, or session.
Session and agent tracing: Multi-step workflows, like an agent that calls a model several times to complete one task, get grouped into a single trace you can inspect end to end.
Caching and rate limiting: The gateway can cache repeated prompts to cut cost and enforce rate limits per user or API key before a request ever reaches the provider.
Prompt versioning: Teams save and compare prompt versions using real production data, then roll out changes through the gateway without redeploying code.
Cost dashboard for a multi-model product: Track exactly how much each customer or feature costs across OpenAI, Anthropic, and open-weight models routed through Helicone, then set budgets per user tier.
Agent debugging tool: Use session tracing to follow a coding agent or support bot through every model call it makes, then find the exact step where a chain of prompts went wrong.
Prompt A/B testing pipeline: Version two prompts, route live traffic between them through the gateway, and compare quality and cost side by side before picking a winner.
Rate-limited internal AI tool: Put a caching and rate-limiting layer in front of an internal chatbot so one team cannot burn through the shared model budget.
Compliance-ready LLM logging: Self-host Helicone under Apache 2.0 to keep every prompt and response inside your own infrastructure for SOC 2 or HIPAA requirements.
Teams use Helicone to monitor and debug LLM applications in production. It logs every request that passes through its proxy, tracking cost, latency, and token usage automatically. Developers use it to trace multi-step agent sessions, version and test prompts, and control spend with caching and rate limiting, all without changing their existing OpenAI or Anthropic code beyond the base URL.
Yes. Helicone's Hobby tier is free and includes 10,000 requests a month with 7-day data retention, no credit card required. The Pro plan costs $79 a month with unlimited seats and longer retention, and Team costs $799 a month with SOC 2 and HIPAA compliance. Enterprise pricing is custom for larger workloads.
Yes. Helicone is open source under the Apache 2.0 license, and every tier, including the free Hobby plan, can be self-hosted. Teams that need full control over their data run Helicone on their own infrastructure, while others use the managed cloud version to skip the setup and maintenance work.
RAGAS is an open-source evaluation framework for RAG systems that scores faithfulness, answer relevancy, context precision, and context recall automatically.
ObservabilityPromptfoo is an open-source CLI and library for testing, evaluating, and red-teaming LLM applications and agents before they reach production.
ObservabilityVendor-neutral standard for traces, metrics, and logs.
Observability
Braintrust is an AI observability and evaluation platform that helps engineering teams trace, test, and monitor LLM applications in production.
ObservabilityUpdates from the AI world — what shipped, what we’re using in production, and what’s worth your attention. Two emails a month, no spam.
