Helicone Review (2026)

Start here: Helicone is in maintenance mode

Before anything else about the product, the fact that decides it. Mintlify acquired Helicone on 3 March 2026, and Helicone is now in maintenance mode. Only security patches, bug fixes and new-model support ship. No new integrations, no new analytics, no roadmap. Mintlify has said it’s helping existing customers migrate off.

For a new buyer, that’s the whole story. It doesn’t matter how clean the product is if the people who own it are winding it down and pointing you toward the exit. We rate Helicone low for exactly one reason: it’s a dead end for new adopters. Everything below is context for existing users deciding when to leave, not a reason to start.

If you’re evaluating observability tools fresh in mid-2026, this is not the one. Skip to Langfuse or Braintrust.

What Helicone actually is (was)

Helicone is the lightest-weight of the major observability tools - a logging proxy first, with evals a distant second. You point your LLM calls at it, and it logs requests, tracks cost, and gives you a dashboard and an AI gateway that can route across providers.

The defining characteristic is the proxy model. You don’t install an SDK and instrument your code. You change your API base URL to route through Helicone, and it logs everything in the middle. That works with any language and any provider - OpenAI, Anthropic, whatever - with zero code changes. There’s also an async SDK logging mode for Python and JS when you don’t want the proxy in your critical path.

It came out of Y Combinator (W23), founded in 2023 in San Francisco by Justin Torre, Cole Gottdank and Scott Nguyen. By the time Mintlify bought it, it had processed 14.2 trillion tokens across roughly 16,000 organizations. Funding reports conflict - some say a $500K YC seed, others around $5M at a $25M valuation - so treat the exact number as unverified.

Pricing: still published, but the meter is a mystery

The tiers are live, but there’s a real gap in what’s disclosed.

Tier	Price	Included	Retention
Hobby	$0	10k requests/mo, 1 GB storage, 1 seat	7 days
Pro	$79/mo	10k free, then usage-based	1 month
Team	$799/mo	10k free, then usage-based	3 months
Enterprise	Custom	SAML SSO, on-prem, custom MSA	Custom

Here’s the catch: the exact per-request overage rate above 10k requests is not published. The page just says “usage-based,” and the $79 Pro price is a floor, not a ceiling. We could not find the actual overage rate from any primary or third-party source, so we’re flagging it as not found. If you’re on Pro at any real volume, you can’t forecast the bill from public information - and now there’s nobody shipping a clearer pricing page.

The free Hobby tier is generous enough for prototyping: 10k requests a month, though only 7-day retention and a single seat.

Self-hosting: free and open, but you’re maintaining a frozen codebase

This is the one genuinely bright spot, with a heavy asterisk.

Helicone is open source under Apache 2.0. You can self-host with Docker or Helm on Kubernetes, no license fee, and the AI gateway is fully open-sourced on GitHub. That’s more open than LangSmith or Braintrust, and on paper it’s a real strength. Earlier Helicone marketing implied self-host was Enterprise-only, but the OSS repo and Apache license contradict that - anyone can run it.

The normal trade-off applies: self-hosting means you own the database, the proxy, ingestion, storage, metadata, the response cache and the scaling path. The subscription savings turn into on-call and infra time.

But the maintenance-mode status changes the calculus. Self-hosting an open-source tool is usually a bet that the upstream project keeps improving and you benefit for free. Here, upstream is frozen. You’d be maintaining a codebase that only gets security and bug fixes, and whose own creator is helping people migrate away. That’s a much weaker bet than it looks.

The architectural gotcha, underneath the acquisition

Even before maintenance mode, the proxy model had a structural cost. Putting Helicone in the request hot path makes it a single point of failure. If Helicone has downtime, your LLM calls fail even when OpenAI and Anthropic are perfectly healthy. And every proxied call carries added latency - Helicone cites around 10ms, but that compounds for real-time apps.

You can dodge both with async SDK logging instead of the proxy, but then you lose the zero-code convenience that was the main reason to pick Helicone in the first place.

Should you use it?

Use Helicone if you’re already on it, you value the lightweight open-source proxy, and you understand you’re now on a frozen product - in which case, use the runway to plan your migration.

Don’t use Helicone if you’re evaluating fresh. The maintenance-mode status makes it a dead end no matter how tidy the tool is, and the vendor itself is pointing customers toward the door.

Bottom line: Helicone was a clean, open, proxy-based observability tool, and in a different timeline it’d be an easy recommendation. But Mintlify put it in maintenance mode in March 2026, and that single fact overrides everything else for a new buyer. If you’re on it, plan your exit. If you’re not, don’t start - go look at Langfuse or Braintrust instead.

---

Pricing and features verified against helicone.ai on 23 July 2026. This category ships breaking changes monthly - we re-verify every 30 days.