Helicone
Also known as: Helicone AI
Open source LLM observability platform and AI gateway that logs, routes, caches, and costs every request through one OpenAI compatible endpoint.
Helicone is an open source platform that pairs LLM observability with an AI gateway, aimed at teams shipping LLM powered products who need to see and control what their applications are doing. The observability side integrates in a single line, by changing a base URL or adding a header, after which every request is logged automatically with its prompt, completion, token counts, cost, latency, and errors, with no wrapper libraries or extra instrumentation. It is a Y Combinator Winter 2023 company, licensed under Apache 2.0, and reports use by more than a thousand AI teams across nearly ten billion requests.
On the observability side, Helicone turns raw traffic into traces and sessions for agents, chatbots, and pipelines, with a dashboard, per user analytics for chargeback, custom metadata for segmenting by feature or customer, and a query language called HQL for digging through logs. It tracks cost using an open source pricing database covering more than three hundred models, and adds caching to cut spend and latency on repeated requests, alerts and scheduled reports for cost and error regressions, a prompt playground, prompt versioning from production data, and experiments. Logs export to tools like PostHog in one line.
The newer AI Gateway, a lightweight Rust service the company calls the NGINX of LLMs, gives one OpenAI compatible endpoint and key that reaches more than a hundred models across providers like OpenAI, Anthropic, Google Vertex, and AWS Bedrock. It routes intelligently to the cheapest, fastest, or most reliable option, falls back automatically when a provider fails, enforces rate limits, and caches responses, with observability built in by default. Pricing through the gateway is zero markup, so you pay exactly what providers charge, and you can bring your own provider keys.
Both the platform and the gateway are open source and can run in Helicone's cloud or be self hosted, including on premises for teams with stricter requirements, and the platform is SOC 2 and GDPR compliant. Pricing is a free Hobby tier covering 10,000 requests a month with one seat, a Pro plan at $79 a month with unlimited seats, alerts, and HQL, a Team plan at $799 a month that adds HIPAA and multiple organizations, and a custom Enterprise tier with SAML single sign on and on premises options. Gateway usage is billed as pass through provider credits at no markup.
Vendor details
Canonical URL
https://www.helicone.ai
Category
Agent infrastructure
Subcategory
Observability and gateway
Funding status
A Y Combinator Winter 2023 company. The observability platform and its AI Gateway are both open source under Apache 2.0. Reports use by more than 1,000 AI teams and processing of nearly 10 billion requests. Independent.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
One line integration by changing the base URL or adding a header, compatible with the OpenAI SDK and working with OpenAI, Anthropic, Google, Azure, AWS Bedrock, LiteLLM, OpenRouter, and Gemini. The open source AI Gateway reaches 100 plus models through one endpoint, and logs export to tools like PostHog.
In practice
You ship an LLM feature with no visibility into cost or failures. You change one base URL to route through Helicone, and every request is logged with its cost, latency, and errors in a queryable dashboard.
You want the cheapest provider per call with automatic failover. Helicone's AI Gateway gives one endpoint to 100 plus models, routes by cost, and fails over to another provider when one is down.
Your AI spend is climbing and finance wants to know which customers drive it. You tag requests with custom properties in Helicone, track per user cost, and set alerts to catch spend spikes early.
Sources & related URLs
Related / legacy domains
Capability coverage
5.5 / 14 capabilities · 39%
| Integrations & Tool CallingOne line integration compatible with the OpenAI SDK and working across OpenAI, Anthropic, Google, Azure, AWS Bedrock, LiteLLM, OpenRouter, and Gemini, plus exports to tools like PostHog, but it is a provider and logging integration layer rather than a tool calling hub. | Partial |
|---|---|
| Workflow OrchestrationThe gateway routes and fails over between model providers at the request level, but does not orchestrate agent workflows, sequencing, or branching. | Unable to verify |
| Knowledge Grounding & RAGObserves and routes LLM traffic but does not provide retrieval or knowledge grounding. | Unable to verify |
| Human Oversight & GuardrailsOffers operational controls like rate limits, but no human review queues or content guardrails on agent outputs. | Unable to verify |
| Security, Identity & GovernanceSOC 2 and GDPR compliant with HIPAA on the Team tier, SAML single sign on and on premises on Enterprise, plus audit logs, data retention controls, and self hosting for data control, though identity governance depth is tier gated. | Partial |
| Observability & AuditabilityCore product. Automatic per request logging of prompts, completions, cost, latency, and errors, with traces and sessions, a dashboard, per user analytics, custom metadata, a query language (HQL), alerts, and reports. | Full |
| Memory & State PersistencePersists logs, traces, and prompt versions, but does not provide an agent memory or state persistence layer. | Unable to verify |
| Deployment & Data ResidencyBoth the observability platform and the AI Gateway are open source under Apache 2.0 and can run in Helicone's cloud or be self hosted, including on premises, giving full deployment and data residency control. | Full |
| Prebuilt Agents, Templates & PacksProvides an open source model pricing database and registry as reference data, but no prebuilt agents, templates, or installable packs. | Unable to verify |
| Triggers & Channel CoverageAlerts, scheduled reports, and webhooks fire on cost and error regressions, but there is no conversational channel coverage or agent invocation runtime. | Partial |
| Model Flexibility & RoutingThe open source AI Gateway is a genuine routing layer: one OpenAI compatible endpoint reaching 100 plus models across 20 plus providers, with cost, speed, and reliability based smart routing, automatic fallbacks, load balancing, and bring your own provider keys. | Full |
| APIs, SDKs & MCP ExtensibilityComprehensive REST and OpenAI compatible APIs, official SDKs including a Vercel AI SDK provider, webhooks, and a fully open source codebase, though no MCP server is documented. | Partial |
| Testing, Debugging & OptimizationA prompt playground, experiments and A/B testing, prompt versioning from production data, trace replay for debugging, and fine tuning integrations, though evaluation is lighter than dedicated eval platforms. | Partial |
| Browser & Computer UseNot applicable. Helicone is an observability and gateway layer and does not provide browser automation or computer use. | Unable to verify |
Pricing
From $79/mo · free tier (10k requests/mo) + open source
Subscription tiers by features and seats, with usage based scaling above included request volume; AI Gateway billed as zero markup provider pass through
Included quota
Hobby free 10,000 requests/mo (1GB storage, 1 seat). Pro $79/mo (unlimited seats). Team $799/mo (5 orgs). Enterprise custom. AI Gateway provider usage is pass through at 0% markup.
What is public
Helicone publishes its full tier structure (Hobby, Pro, Team, Enterprise) with request, seat, and feature limits, plus the zero markup gateway model. Enterprise pricing is custom.
Billing mechanics
A feature and seat based subscription on the observability platform, with Pro scaling on request volume. Routing model traffic through the AI Gateway is billed as pass through provider credits at no markup. The open source builds carry no license fee when self hosted.
Cost watchouts
Request volume drives Pro scaling, and gateway provider pass through grows with model usage. Compliance (HIPAA) and SSO sit on higher tiers.
Variable cost rationale
The subscription tiers are predictable, but Pro scales with request volume above the included amount, and if you route model traffic through the AI Gateway you also pay pass through provider costs that grow directly with usage, albeit at zero markup.
Additional watchouts
Two cost layers can stack: the platform subscription and, if you use the gateway, pass through provider spend. HIPAA is gated to the Team tier and above, and SAML SSO and on premises to Enterprise.
Overage / add-ons
Pro scales with usage above the included request volume. AI Gateway usage is billed as pass through provider credits at zero markup, so you pay exactly what providers charge plus the Stripe processing fee. Gateway credits are added at helicone.ai/credits.
Sales call required
No — self-serve available
Free / trial
Free Hobby tier: 10,000 requests/month, 1GB storage, 1 seat, no card. Open source and self hostable.
Lowest paid plan
Pro $79/mo (unlimited seats, alerts, HQL)
Commercial notes
Open source led, bottom up adoption through a generous free tier and one line integration, with Pro and Team as self serve upgrades and Enterprise for SSO, on premises, and compliance. Strong fit for teams running multi provider LLM traffic.
Key ambiguities
How Pro request overage is priced above the included volume, and where Enterprise lands, are not fully itemized.
Cancellation / refund
Hobby, Pro, and Team are self serve subscriptions with standard cancellation. Enterprise terms are contractual.
Support SLA / resale
Community support on Hobby, standard support on Pro, Slack support on Team, and dedicated support with an SLA on Enterprise.
Missing data
Enterprise pricing is custom. Exact request overage rates on Pro above the included volume are not itemized.
Related vendors
- AgentOps — Agent observability and reliability platform with broad model and…
- Agno — High-performance agent runtime and framework (formerly Phidata) with…
- Apify — Cloud platform for web scraping and automation with 45,000+ prebuilt…
- Arcade — Authenticated tool calling platform and MCP runtime that handles…
- Arize AI — AI observability and evaluation platform that traces, evaluates, and…
- Braintrust — AI evaluation and observability platform with self-serve pricing,…