Helicone Alternatives: 9 Picks for Proxy-Style LLM Logging

The best Helicone alternatives depend on which part of Helicone you'd miss. If you want to keep the one-line proxy setup, look at Portkey, the LiteLLM proxy or Cloudflare AI Gateway. If the proxy was only there for logging, move to an SDK-based tool: Langfuse for open-source tracing, Arize Phoenix for the lightest self-host, PostHog if you want LLM calls next to product analytics, or LangSmith if you build on LangChain. Braintrust sits in between, and OpenLLMetry is the pick if you don't want any one vendor to own your instrumentation.

Helicone's appeal was always simple. Change one base URL, and every model call gets logged, costed and optionally cached. Most alternatives make you choose between that proxy shape and deeper tracing. This guide starts with what's happening at Helicone, then sorts nine options by the three things Helicone users rely on: proxy integration, caching and cost tracking.

What's Happening at Helicone in 2026

Here's what the primary sources say, in order.

March 3, 2026. Mintlify announced it was acquiring Helicone, with Helicone's co-founders joining Mintlify in San Francisco. The post said: "Helicone will continue operating in maintenance mode—security updates, bug fixes, and new models will keep shipping." It also said Mintlify would "support a smooth migration to another platform" for every customer. It gave no shutdown date and named no replacement.

Now. Helicone's own site carries a banner: "Helicone is no longer part of Mintlify. Details to come soon." Neither company has published those details. The post Helicone wrote about joining Mintlify is no longer on its blog.

What still works. When we checked on November 12, 2026:

• Helicone's pricing page still listed plans and a sign-up button.
• The GitHub repository was public, Apache-2.0 and not archived. Its most recent commit was dated October 28, 2026.
• The docs were online, with no deprecation notice on the pages we read.

So nobody has announced a shutdown. But the last official word on the roadmap was "maintenance mode," and the ownership question is open. That's fine for an app already in production. It's a weak basis for choosing a tool you'll depend on for two years.

What Helicone users actually need

Any replacement has to cover some of these. Decide which ones matter before you compare.

One note if you plan to self-host Helicone as your exit plan: its Docker guide says the main API service no longer proxies LLM traffic, so you deploy the AI Gateway alongside it. Test that setup before you rely on it.

Quick Comparison

Proxy-Style Helicone Alternatives

These keep Helicone's shape: your app talks to a gateway, and the gateway logs, caches and tracks spend.

Portkey: the closest feature match

Best for: teams that want logging, caching and cost controls from one gateway.

Portkey's observability docs list request logs, tracing, an analytics view with 21+ metrics, custom metadata, feedback and budget limits on provider keys. Its cache docs describe two modes. Simple cache returns a stored response for an identical request and works on all plans. Semantic cache also matches similar requests, and "is only available on select Enterprise plans."

The gateway itself is open source under MIT, so you can run it yourself.

What to know: Palo Alto Networks closed its acquisition of Portkey on May 29, 2026, and Portkey's site now says it is Prisma AIRS AI Gateway. If you're leaving Helicone because of ownership uncertainty, weigh that. You'd be moving to a product inside a large security vendor, which is stable but aimed at enterprises.

LiteLLM proxy: the one you run yourself

Best for: teams that want the proxy inside their own network and their logs wherever they choose.

The LiteLLM proxy gives you one OpenAI-compatible endpoint, virtual keys and spend tracking per project or user. Two features map directly to Helicone habits:

• Caching. The caching docs list Redis, S3, GCS, in-memory and disk backends for exact-match caching, plus semantic caching on Redis or Qdrant. You turn it on with a cache key in the config file.
• Logging anywhere. Callbacks send every request to Langfuse, OpenTelemetry, Datadog, LangSmith, MLflow, S3 and more.

That second point is why many Helicone leavers pair LiteLLM with a tracing tool. You keep the proxy and pick your own dashboard.

What to know: two malicious LiteLLM releases were uploaded to PyPI on March 24, 2026. LiteLLM's security update says the official Docker image was not affected and a clean release followed on March 30. Our LiteLLM alternatives guide covers the details and the other gateways in this category.

Cloudflare AI Gateway: no servers, free core features

Best for: small teams that want logs, cost numbers and caching without running anything.

Cloudflare AI Gateway is available on all Cloudflare plans. It shows requests, tokens and cost, keeps logs, and adds rate limiting, retries and model fallback. Its caching works much like Helicone's: identical requests are served from Cloudflare's cache, with per-request headers to set the lifetime or skip the cache.

What to know: caching is exact-match only. The docs say semantic caching is planned, not shipped. There's also no self-hosted version, and it doesn't offer evals or prompt management.

Braintrust: a gateway attached to an eval platform

Best for: teams that want proxy-style logging now and serious evaluation later.

The Braintrust Gateway works the Helicone way. Point your SDK at the gateway URL and you get caching, logging and multi-provider access. It has three cache modes, and a header sends each request to a Braintrust project as a trace. Another header sets fallback providers.

What to know: the docs describe the hosted Gateway as a public preview with no service-level agreement, and warn that behaviour may change. Braintrust's older AI proxy is deprecated in favour of it. Self-hosting guides are for organisations that already run a self-hosted data plane; new ones are pointed to bring-your-own-cloud.

SDK-Based Helicone Alternatives

These don't sit in the request path. Your app calls the model provider directly and sends trace data on the side. You lose gateway caching, and you gain richer traces of multi-step agents. You also remove one hop that can fail.

Langfuse: open-source tracing with a migration guide

Best for: teams that used Helicone mainly as a log viewer and want evals and prompt management too.

Langfuse publishes its own guide to migrating from Helicone. It's honest about the gap: Langfuse captures calls in your application through SDKs or OpenTelemetry, not at a gateway. Its advice is that if the gateway was only there for logging, drop it, and if you still want one, put LiteLLM's proxy in front.

Cost tracking carries over. Langfuse records usage and cost per generation, either sent from the model response or calculated from usage and model prices. Response caching doesn't. That was a Helicone gateway feature.

Self-hosting is free. For the Helicone vs Langfuse question: Helicone is faster to adopt, and Langfuse goes deeper once you have agents, evals and prompts to manage. Langfuse joined ClickHouse in January 2026 and said no licensing changes are planned. Our Langfuse alternatives roundup covers that.

Arize Phoenix: the lightest self-host

Best for: teams that want a self-hosted trace viewer running in minutes.

Phoenix installs with pip and starts with phoenix serve. It's built on OpenTelemetry, with auto-instrumentation from the OpenInference project, and covers tracing, evals, versioned datasets, experiments and a prompt playground.

What to know: the licence is Elastic License 2.0, which is source-available, not OSI open source. Phoenix has no gateway and no response cache.

PostHog: LLM calls next to product analytics

Best for: product teams already using PostHog.

PostHog's AI observability captures prompts, responses, tokens, cost, latency and tool calls, then links each trace to the user, their session replay and any exceptions. You install it by wrapping your provider client with PostHog's SDK, and the wrapped client records a generation event with cost and token properties. It's close to Helicone's low-effort setup, without a proxy.

What to know: it isn't a gateway, so there's no caching or routing. PostHog offers an MIT-licensed Docker Compose deployment, but its docs say self-hosted instances get no support or guarantees.

LangSmith: if you build on LangChain

Best for: teams on LangChain or LangGraph who want tracing, evals and agent hosting from one vendor.

LangSmith traces through its SDK or OpenTelemetry. It's also adding the piece Helicone users know: an LLM Gateway, in beta, that routes calls to providers, applies fallbacks, spend limits and rate limits, and turns each call into a trace. The docs don't mention response caching.

What to know: LangChain's pricing page bills per seat plus usage, and self-hosting is an Enterprise option. Our LangSmith alternatives post covers who that suits.

OpenLLMetry: instrumentation nobody owns

Best for: teams that never want to redo instrumentation when a vendor changes hands again.

OpenLLMetry is an Apache-2.0 set of OpenTelemetry extensions for LLM providers and vector databases. It emits standard spans, so the backend is a config choice: Langfuse, Phoenix, Datadog, or whatever comes next.

What to know: it's a library, not a dashboard. And its maker, Traceloop, is joining ServiceNow. The announcement says OpenLLMetry "will remain open-source."

How to Choose a Helicone Alternative

Three practical tips before you move:

1. Export your data first. Whatever Helicone's future, get your request history out while the service is live.
2. Separate the two jobs. Routing and caching belong in a gateway. Tracing and evals belong in an observability tool. Helicone did both; your replacement can be two tools.
3. Count the acquisitions. Helicone, Portkey, Langfuse and Traceloop all changed owner or announced a deal in 2026. Standard formats, the OpenAI API shape for requests and OpenTelemetry for traces, are your insurance.

Where Taku Fits

Taku isn't an observability tool or a gateway, so it doesn't replace Helicone or anything else on this list. Everything above is for teams who build LLM apps and need to see inside them.

Taku is for the other side of that work. Taku Apps and Sites, built by the best AI minds. Find one, run it, and keep it in your Mac Dock. Taku is in Beta, and the Mac app is available now.

FAQ

What is the best Helicone alternative?

If you want to keep proxy-style logging, Portkey is the closest feature match and the LiteLLM proxy is the closest self-hosted one. If you only used Helicone to view logs and costs, Langfuse is the most common move.

Is Helicone shutting down?

No shutdown has been announced. Mintlify said in March 2026 that Helicone would run in maintenance mode, and Helicone's site now says it is no longer part of Mintlify, with details to come. Treat the roadmap as uncertain until those details are published.

Is there an open-source Helicone alternative?

Yes. For the proxy, LiteLLM (MIT core) and the Portkey gateway (MIT). For tracing, Langfuse (MIT core) and OpenLLMetry (Apache-2.0). Phoenix is free to self-host under the source-available Elastic License 2.0. Helicone's own code is also Apache-2.0.

Helicone vs Langfuse: what's the difference?

Helicone logs calls at a gateway in the request path. Langfuse logs them from inside your app through SDKs or OpenTelemetry. Helicone adds caching and routing. Langfuse adds deeper agent tracing, evals and prompt management.

Which Helicone alternatives support response caching?

LiteLLM, Portkey, Cloudflare AI Gateway and the Braintrust Gateway. Cloudflare's is exact-match only. SDK-based tools such as Langfuse, Phoenix and PostHog don't cache model responses.

Can I keep using Helicone?

Yes, for now. The cloud service, docs and Apache-2.0 repository were all still available on November 12, 2026. Keep an export of your data and a tested fallback either way.