Is Helicone open source?

Partly, under two licenses: the main platform in the Helicone/helicone repository is Apache-2.0, while the separate Rust-based AI Gateway in the Helicone/ai-gateway repository is GPL-3.0 according to the GitHub license field and is marked Public Beta there (as of 26 September 2026). A hosted offering exists alongside, with Team and Enterprise tiers among others.

What does the acquisition of Helicone by Mintlify mean?

According to the Helicone blog of March 2026, Helicone has been acquired by Mintlify. The post states that security updates, new models and bug and performance fixes continue to ship. The vendor's current information always prevails.

Does Helicone measure the working time of coding agents?

According to its documentation (as of 26 September 2026), Helicone records what passes through the proxy: requests, tokens, cost. Sessions group related requests into a trace but are not a time measurement. The innFactory AI Gateway companion reads the local session logs of Claude Code, Codex and OpenCode and reports agent time and subscription usage per repository.

Can I set budgets in EUR per cost center?

Helicone documents cost-based rate limits in US cents per API key, user or custom property with a hard stop via HTTP 429; a EUR display is not documented. The innFactory AI Gateway runs cost centers with pool or per-user budgets, per-member caps and EUR display at the ECB daily rate.

Helicone Alternative: innFactory AI Gateway Compared

Key facts (as of 26 September 2026):
Helicone describes itself as an “AI Gateway & LLM Observability” platform, available as a hosted cloud with EU or US region choice or via self-hosting (Docker Compose, Helm chart); the main platform is Apache-2.0, the separate Rust AI Gateway is GPL-3.0 according to the GitHub license field and marked Public Beta, and according to the Helicone blog of March 2026 Helicone has been acquired by Mintlify.
The innFactory AI Gateway is deployed into the customer’s cloud tenant and bundles cost centers with per-member caps, budgets in euros and a companion that measures the agent time of Claude Code, Codex and OpenCode per repository, including subscription usage.
Per its documentation (as of 26 September 2026), Helicone records what passes through the proxy (requests, tokens, cost), controls cost through rate-limit policies in US cents with a hard stop via HTTP 429 and groups requests into sessions as traces, which are not a time measurement; the innFactory AI Gateway companion reads the local session logs of Claude Code, Codex and OpenCode and reports agent time and subscription usage per repository, displayed in euros.
Helicone fits when you develop your own AI applications and primarily need tracing, debugging and sessions; the innFactory AI Gateway fits when you want to steer the AI usage of an organisation – cost centers in EUR, roles from Entra ID or Keycloak, guardrails across all models and agent time per repository in your own tenant.

Helicone describes itself as an “AI Gateway & LLM Observability” platform for AI engineers (as of 26 September 2026). The product comes from the observability side: every request to a language model is logged, tagged with metadata, grouped into sessions and analysed in a dashboard. For teams building their own AI applications that is a strong tool — Helicone is a good product.

The innFactory AI Gateway comes from the other direction: an infrastructure component in the customer’s cloud tenant that puts language models and coding agents under cost centers, budgets, roles and guardrails, natively integrated with CompanyGPT.

The real decision is therefore: do you primarily need request tracing for AI applications you build yourself — or cost control, identity and governance for the AI usage of an organisation?

What Helicone is

Helicone, Inc. was started, in its own words, “three years ago during YC W23”. No registered office is stated on the pages we reviewed; the terms refer to California law, the privacy policy names the United States as the place of data processing.

Two repositories, two licenses: the main platform (github.com/Helicone/helicone) is Apache-2.0. The separate AI Gateway written in Rust (github.com/Helicone/ai-gateway) is GPL-3.0 according to the GitHub license field and the LICENSE file, and is marked “Public Beta”. According to the pricing page, the hosted offering includes the Team and Enterprise tiers among others; the README additionally mentions a free tier.

According to the Helicone blog of March 2026, Helicone has been acquired by Mintlify (“Helicone has been acquired by Mintlify, and our team will be joining them in San Francisco”). In its own wording, the post states that “security updates, new models, bug & performance fixes all keep shipping”.

Operating model: SaaS, self-hosting or infrastructure in your own tenant

Helicone offers a hosted cloud and self-hosting with four documented paths: “Manual Installation”, “Docker Compose”, “Kubernetes” via Helm chart, and Cloud. Per the guide, the self-hosted stack consists of Postgres, ClickHouse, Minio, Jawn and the web frontend; as “Jawn no longer proxies LLM traffic”, the separate AI Gateway is added for proxying. In the cloud, Helicone lets you choose between EU and US regions according to its documentation and provides a SOC 2 report on request.

The innFactory AI Gateway is not hosted by innFactory. It is deployed as a container into the customer’s cloud tenant — Azure Container Apps, AKS, STACKIT or any Kubernetes — and operated there as part of the customer’s own AI stack; gateway, dashboard and data stay in the tenant; innFactory is not in the request path. Configuration is YAML, a Terraform provider or natural language via the built-in gateway-config MCP server. The current version is 1.0.0-rc16, in production with first customers since August 2026.

Identity, SSO and roles

According to the pricing page, “SAML SSO” is included from the Team tier upwards. Named identity providers, SCIM and roles are not documented on the pages we reviewed. End users are identified via the Helicone-User-Id header, which per the documentation enables per-user cost and usage metrics; per-user model allow-listing is not documented.

With the innFactory AI Gateway, identity is part of every license tier: Microsoft Entra ID or generic OIDC (Keycloak documented). Seven roles — Owner, Admin, Auditor, Finance, Security, Member, Guest — are included; six of them map from IdP groups, Guest is assigned automatically for logins outside the allowed e-mail domains; cost centers have their own administrators, and models, budgets and rate limits can be set per person.

Cost control and AI FinOps

Helicone’s cost control works through “Custom Rate Limits”: a policy in the header Helicone-RateLimit-Policy: [quota];w=[time_window];u=[unit];s=[segment], applied across the whole API key, per user or per custom property. The unit is request or cents. When the limit is reached, Helicone responds with HTTP 429 per its documentation — a hard stop. Attribution uses custom properties, which let cost be calculated “per user, conversation, or feature”. The documented examples work in US cents; a EUR display is not documented.

The innFactory AI Gateway has four independent gates: organisation, cost center (monthly plus daily and weekly windows, pool or per-user mode with a cap per member), user and API key. An exhausted gate responds with HTTP 402 and error code E4006 — optionally a budget fallback model takes over, while per-member caps always remain hard stops. Cost is attributed by user, key, team, cost center, model, provider, tool, repository and billing source, and shown in EUR at the ECB daily rate; prices come from ai-prices.eu with admin review before they apply. Every buffered response carries the x-aigateway-response-cost header; Insights and the CSV export also include agent and LLM time.

Coding agents, subscriptions and agent time

For Claude Code, Helicone documents an integration via base URL: ANTHROPIC_BASE_URL points to the Helicone endpoint, after which “any calls to Claude Code” are logged. Per the same page, this method is “maintained but no longer actively developed”; the vendor points to its new AI Gateway. Measured is what passes through the proxy: requests, tokens, cost. According to Helicone’s documentation (as of 26 September 2026), a local companion that records agent working time or subscription usage independently of gateway traffic is not described. Helicone’s “Sessions” are something else: they group requests into a trace (“trace the entire agent flow from initial user input to final response in one unified view”) — a debugging tool, not a time measurement.

The innFactory AI Gateway provides two building blocks. First, proxy setups for six tools — Claude Code, Codex, OpenCode, Cowork, VS Code Continue and VS Code Copilot (BYOK) — each developer with an own key and cost center. Second, the aigw companion on the developer machine (Linux, macOS, Windows), which reads the local session logs of Claude Code, Codex and OpenCode and reports agent time and LLM time per repository, user, tool and cost center. It detects whether a session ran via the gateway or the tool’s subscription (Claude Max, Codex plan); subscription usage is valued at list price, never billed or budgeted. Rollout runs via MDM through Intune, a macOS policy or a Linux script. To be candid: the companion does not cover Cursor, Windsurf, Aider, Cline, Copilot CLI or Gemini CLI, and the binaries are not code-signed yet.

Providers, dialects and models

Helicone states “100+ models and 20+ providers”; its table of approved domains lists around two dozen providers, including OpenAI, Anthropic, Google Gemini and Mistral. The primary inbound dialect is OpenAI-compatible, the Anthropic API is passed through natively, and there is an SDK example for Vertex AI. We did not find the Responses API, realtime or audio on the pages reviewed; STACKIT, IONOS and Aleph Alpha are not documented as providers. In provider breadth, Helicone is ahead.

The innFactory AI Gateway connects 13 provider types: OpenAI, Azure OpenAI, Azure AI Foundry, Anthropic (direct or Vertex), Gemini (AI Studio or Vertex including the EU endpoint), Mistral, AWS Bedrock, STACKIT, IONOS, Ollama, LM Studio, ElevenLabs and OpenAI-compatible endpoints. Inbound it speaks three dialects — OpenAI, Anthropic and Gemini — plus realtime WebSockets, audio, embeddings and images.

Governance: guardrails, MCP, routing and Flow Builder

Guardrails. Helicone documents “Moderations” — a gate on OpenAI’s moderation endpoint; flagged prompts are rejected with HTTP 400, and per the documentation “the feature currently supports OpenAI models only”. PII redaction, a prompt-injection filter or a shadow mode are not documented. The innFactory AI Gateway ships nine rule types — among them secret detection, PII redaction and prompt injection — with shadow mode, mid-stream redaction and a GDPR preset.

MCP. Per its documentation, Helicone provides its own MCP server through which MCP clients query Helicone data. We did not find a proxy for third-party MCP servers. The innFactory AI Gateway runs an MCP proxy with eight upstream auth modes, access policies, toolsets, guardrails on tool arguments and an audit trail per call.

Routing and flows. Helicone documents prioritised fallbacks via the Helicone-Fallbacks header; the Rust gateway’s README mentions “Smart routing with latency and cost optimization”. The cache is exact-match over URL, body and headers. A visual flow builder is not documented. The innFactory AI Gateway routes via an auto-router by prompt complexity, through aliases, fallbacks and deployment groups. The Flow Builder is a visual graph with nine strategies, guardrail nodes with pass/block/fail branches and event triggers to webhooks; a flow is callable like a model and bills the same budgets.

What Helicone does better

Request tracing as the core product, with sessions as traces of entire agent flows.
User metrics and free-form attribution via Helicone-User-Id and custom properties.
Cost-based rate limits with a hard 429 in a single header line.
Alerts on error rate, cost, latency and tokens via email and Slack.
EU/US region choice in the cloud, SOC 2 report on request.
Four self-hosting paths including a Helm chart, main platform under Apache-2.0.
Broader provider list and its own MCP server for observability data.

When Helicone is the right choice

You develop your own AI applications and primarily need tracing, debugging and sessions.
Your application sets the attribution tags itself.
A hosted service with an EU region is sufficient, or you run the stack yourself.
Integration should be a header swap in code.

When the innFactory AI Gateway is the right choice

The data path must stay in your own cloud tenant — Azure, STACKIT or your own Kubernetes.
You want budgets in EUR per cost center, member and key, with hard stop or fallback model.
Entra ID or Keycloak with roles should work in every license tier.
Coding agents partly run on subscriptions, and you want agent time per repository.
Guardrails, MCP registry and Flow Builder should be the same ones CompanyGPT uses.

Comparison at a glance

Criterion	Helicone (per documentation, as of 26 Sep 2026)	innFactory AI Gateway
License	Platform Apache-2.0; AI Gateway GPL-3.0, Public Beta	Commercial, EUR 199 to 749 per month
Operation	Cloud (EU/US) or self-hosting incl. Helm chart	Container in customer tenant: Azure, STACKIT, Kubernetes
Identity	SAML SSO from Team tier; roles not documented	Entra ID/OIDC, seven roles, every license
Budgets	Rate-limit policy per key, user, property in US cents; HTTP 429	Gates organisation, cost center, user, key; 402 or fallback model; EUR at ECB rate
Attribution	Custom properties, Helicone-User-Id	User, cost center, model, tool, repository, billing source
Coding agents	Claude Code via base URL; measures proxied requests	Six proxy setups; companion measures agent time and subscription usage
Providers	100+ models, 20+ providers	13 provider types incl. STACKIT, IONOS, Bedrock
Guardrails	OpenAI moderations, OpenAI models only	Nine rule types, shadow mode
MCP	Own MCP server for Helicone data	MCP proxy with eight auth modes
Routing	Prioritised fallbacks; exact-match cache	Auto-router, Flow Builder with nine strategies

Conclusion

Helicone is an observability product with a gateway; the innFactory AI Gateway is a governance and FinOps component with observability. If you build your own AI applications and want to see what every agent step costs, Helicone offers mature tracing with little integration effort.

If you want to steer the AI usage of an organisation — cost centers in EUR, roles from the directory, guardrails across all models, coding agents including subscription usage, all in your own tenant — the innFactory AI Gateway brings those building blocks together.

Natively integrated with CompanyGPT, one reusable stack

The innFactory AI Gateway is natively integrated with CompanyGPT: the same budgets, cost centers and guardrails apply to chat, agents and add-ons. Identity via Entra ID or Keycloak, cost centers, guardrails and the MCP registry form one stack reused across CompanyGPT, the gateway and the customer’s own tools. The gateway can equally run standalone. Further comparisons are listed under all comparisons; the product page is AI Gateway.

Sources

Retrieved on 26 September 2026:

Main repository (Apache-2.0) — https://github.com/Helicone/helicone
AI Gateway repository (Public Beta) — https://github.com/Helicone/ai-gateway
GitHub API, main repository — https://api.github.com/repos/Helicone/helicone
GitHub API, AI Gateway — https://api.github.com/repos/Helicone/ai-gateway
AI Gateway LICENSE (GPL-3.0) — https://raw.githubusercontent.com/Helicone/ai-gateway/main/LICENSE
Blog: Joining Mintlify — https://www.helicone.ai/blog/joining-mintlify
Terms of Service — https://www.helicone.ai/terms
Privacy Policy — https://www.helicone.ai/privacy
Pricing (tiers, SAML SSO) — https://www.helicone.ai/pricing
Self-hosting overview — https://docs.helicone.ai/getting-started/self-host/overview
Self-hosting, manual — https://docs.helicone.ai/getting-started/self-host/manual
Data Autonomy (EU/US, SOC 2) — https://docs.helicone.ai/references/data-autonomy
FAQ Security — https://docs.helicone.ai/faq/tag/security
Custom Rate Limits — https://docs.helicone.ai/features/advanced-usage/custom-rate-limits
Custom Properties — https://docs.helicone.ai/features/advanced-usage/custom-properties
Claude Code integration — https://docs.helicone.ai/integrations/anthropic/claude-code
Gateway integration — https://docs.helicone.ai/getting-started/integration-method/gateway
Gateway fallbacks — https://docs.helicone.ai/getting-started/integration-method/gateway-fallbacks
Moderations — https://docs.helicone.ai/features/advanced-usage/moderations
Caching — https://docs.helicone.ai/features/advanced-usage/caching
Sessions — https://docs.helicone.ai/features/sessions
Alerts — https://docs.helicone.ai/features/alerts
Helicone MCP server — https://docs.helicone.ai/integrations/tools/mcp

Note on the information: All statements about other vendors’ products are based on their publicly available documentation as of the stated date. Vendors continuously develop their products, features and terms — the vendor’s current information always prevails. If any statement appears outdated or incorrect to you, please drop us a line at info@innfactory.ai; we will review and correct it promptly. This comparison does not replace legal or data-protection advice in individual cases.

AI Gateway vs. Helicone: Observability platform with a gateway, or a FinOps component in your own tenant?