Helicone

Helicone is an Apache-2.0 LLM observability platform, self-hostable and sold as a hosted service in US and EU regions metered in requests.

Last verified

Assessed attributes for Helicone, each with the source it came from and the date it was checked
Licence Apache-2.0 — the vendor states Helicone is licensed under the Apache License 2.0; the hosted cloud product shares the Helicone/helicone code base (LICENSE file at tag v2025.08.21-1 is Apache 2.0) and is sold as a paid service on the pricing page. Mintlify acquired Helicone (vendor blog post, 3 March 2026, https://www.helicone.ai/blog/joining-mintlify, checked 22 August 2026); the licence itself is unchanged. Source, checked
Self-host yes — self-host documented via Manual installation, Docker Compose, Kubernetes (Helm) or cloud deployment; vendor-supported "On-prem deployment" is listed only on the Enterprise plan Source, checked
OpenTelemetry unstated for the hosted platform — docs.helicone.ai documents no OTLP ingest endpoint (llms.txt contains zero matches for OpenTelemetry/OTLP/OTel, checked 2026-08-22); the nearest documented route is an OpenLLMetry async-logging integration, which the vendor does not describe as OTel/OTLP ingest. (Separate product: the Helicone AI Gateway, a differently named repo Helicone/ai-gateway, claims "OpenTelemetry support for logs, metrics, and traces" in its own README — that is the gateway emitting its own telemetry, not OTLP ingest into the hosted platform.) Source, checked
Free tier limits Hobby (Free): 10,000 requests/mo, 1 GB storage, 1 seat, 1 organization; 7 days retention; 10 logs/min ingestion; 10 API calls/min Source, checked
Price per 100k traces unstated — Helicone prices in requests, not traces: Pro is $79/mo (USD) with "10K free + usage-based" requests, but the per-request overage rate is not published anywhere on the pricing page (only an interactive calculator), so a 100k-request price cannot be derived from published figures Source, checked
Data residency Cloud: choice of EU and US regions (stated as meeting data residency requirements); no other regions stated Source, checked
Retention Hobby 7 days; Pro 1 month; Team 3 months; Enterprise Forever ("Configurable retention" is ticked for Team and Enterprise) Source, checked
SDK languages First-party manual-logger SDKs in TypeScript/Node.js, Python and Go, plus cURL/HTTP; docs additionally list framework integrations (LangChain, LlamaIndex, Vercel AI SDK, OpenAI Agents, Claude Agent SDK, LiteLLM), documented under the AI Gateway section of the same docs site Source, checked
Last release Vendor changelog (the hosted product's own update feed) newest entry dated November 26, 2025 — "Claude Sonnet 4 and Sonnet 4.5 now support 1M context window". The Helicone/helicone repo's latest release tag is v2025.08.21-1 (tag resolves: raw.githubusercontent.com/Helicone/helicone/v2025.08.21-1/LICENSE returns HTTP 200, checked 2026-08-22); that tag string encodes 2025-08-21, one day outside a 12-month window from the assessment date, so freshness for 2.3 rests on the 26 Nov 2025 changelog entry, not on the repo tag. github.com/Helicone/helicone/releases returns HTTP 403 to this session, so the releases page itself is not verifiable here. Source, checked

Assessment

The vendor states Helicone is licensed under the Apache License 2.0, and the hosted cloud shares that code base. Self-hosting is documented four ways (manual install, Docker Compose, Helm, or a cloud deployment); vendor-supported on-premises deployment is Enterprise-only.

OpenTelemetry support is unstated for the hosted platform: the docs describe no OTLP ingest endpoint and mention OpenTelemetry, OTLP or OTel nowhere (22 August 2026). The nearest documented route is an OpenLLMetry integration that logs traces to Helicone, which the vendor does not call OTLP ingest.

Metering is in requests, not traces or spans. The free Hobby tier gives 10,000 requests a month, 1 GB storage, one seat, 7 days retention and 10 logs a minute. Pro is $79 a month with 10k requests included plus usage-based charging, but the overage rate appears only in an interactive calculator, so no 100k-request price is derivable. Retention runs 7 days to forever across the four plans, in an EU or US region.

The vendor's blog post of 3 March 2026 states "Helicone has been acquired by Mintlify" and that "Helicone's services will remain live for the foreseeable future in maintenance mode. This means security updates, new models, bug & performance fixes all keep shipping" (helicone.ai/blog/joining-mintlify, checked 22 August 2026). A site-wide banner links it from the home page; the Apache-2.0 licence is unchanged.

Something wrong on this page? Suggest a correction — corrections are free and open to everyone.