Helicone is an Apache-2.0 request logger and AI gateway that has received only patches and model support since March 2026. For a new deployment, LangSmith is usually the practical choice, since Helicone is not taking new cloud sign-ups; existing Helicone users can keep a self-hosted copy running while they plan a move.[3][4]
The main difference is status: Helicone has been in maintenance mode since March 2026, with new cloud sign-ups disabled, while LangSmith is LangChain's actively developed platform for tracing and evaluating agents. Mintlify acquired Helicone in March 2026, and on 19 October 2026 Helicone said it is no longer part of Mintlify, with no new owner named. For a new deployment, LangSmith is usually the practical choice; existing Helicone users should plan a migration.
| Decision factor | Helicone | LangSmith |
|---|---|---|
| Status | Maintenance mode since March 2026; new cloud sign-ups disabled | Active development; Engine v2 and Trajectory view shipped September 2026 |
| Ownership | Unclear: says it left Mintlify on 19 October 2026, no new owner named | LangChain, independent; $125M Series B in October 2025 |
| How it captures data | Proxy or AI Gateway in the request path; async SDKs also available | Async SDK or OpenTelemetry, out of the request path; optional gateway |
| Agent tracing | Header-based session grouping, not a span model | Nested runs, threads and a Trajectory view across subagents |
| Evaluation | Not an evaluation framework, by its own docs; Python evaluators turned off October 2026 | Offline and online evals with LLM-as-judge, pairwise and code evaluators |
| Licence and self-hosting | Apache-2.0; free to self-host with Docker or Helm | Proprietary; self-hosting is an Enterprise add-on |
| Pricing | Hobby free, Pro $79 a month, Team $799 a month; new accounts cannot sign up | Developer free for 1 seat; Plus $39 per seat a month |
| Capability | Helicone | LangSmith | Prefactor |
|---|---|---|---|
| Observe | |||
| LLM call tracing | ✓ | ✓ | ✓ |
| Multi-step agent tracing (tool calls, nested spans) | Partial | ✓ | ✓ |
| Cost tracking | ✓ | ✓ | ✓ |
| OpenTelemetry support | Partial | ✓ | ✓ |
| Evaluate | |||
| Offline evals (datasets, experiments) | Partial | ✓ | ✓ |
| Online evals on production traffic | Partial | ✓ | ✓ |
| LLM-as-a-judge scoring | Partial | ✓ | ✓ |
| Code-based evaluators | — | ✓ | ✓ |
| Human review and annotation | Partial | ✓ | ✓ |
| Prompt versioning and playground | ✓ | ✓ | — |
| Act on findings | |||
| Runtime controls on live agent actions | Partial | Partial | ✓ |
| Gateway or proxy in the request path | ✓ | — | — |
| Licence and deployment | |||
| Open source | ✓ | Partial | Partial |
| Self-hosting | ✓ | ✓ | Partial |
Mintlify announced its acquisition of Helicone on 3 March 2026 and moved the product to maintenance: vulnerability patches, bug fixes and support for new models, with help for customers to migrate elsewhere. The cloud sign-up page has shown that sign-ups are disabled since the same day, and the last feature entry in Helicone's changelog is from November 2025. On 19 October 2026 Helicone's site began saying it is no longer part of Mintlify, with details to come and no new owner named.
Work still lands in the open-source repository. In October 2026 a batch of fixes hardened self-hosted Helicone, Python evaluators were turned off, and calls to its judge models were rate limited. In May 2026 an AWS account lock took the dashboard and log processing down for about four days while the proxy stayed up. For a team choosing a tool today, a product with no new sign-ups and an unnamed owner is hard to build on, which is why LangSmith, or another actively developed platform, is usually the practical choice.
Helicone was built as a gateway. You change your provider base URL and it logs every request, with cost per request, caching, failover, rate limits and a prompt registry. Its own documentation says it is not an evaluation framework. Agent runs are grouped through session headers rather than a span model, and its prompt-injection blocking works with OpenAI models only.
LangSmith traces out of the request path through its SDK or OpenTelemetry, models agents as nested runs and threads, and covers the evaluation loop Helicone leaves to other tools: datasets, experiments, online evaluators, annotation queues and Engine. Its LLM Gateway, in public beta since 30 July 2026, is optional and adds spend caps, model access policies and redaction to model calls; it does not cover tool calls.
Helicone's strengths were cost visibility and routing: exact per-request cost through its model registry, pricing for over 300 models, cost alerts and a single endpoint for more than 100 providers. LangSmith calculates cost automatically for most OpenAI, Anthropic and Gemini models and accepts manual costs for others, including tool calls.
Nothing forces an immediate move. The main repository is Apache-2.0, so a self-hosted copy keeps working, and the cloud service still runs for existing accounts. Note that the self-hosted Docker image no longer proxies LLM traffic on its own; the separate AI Gateway, GPL-3.0 and last updated in November 2025, has to run beside it.
To move tracing to LangSmith, two routes exist. Helicone's async logging is built on OpenLLMetry, and LangSmith's OTLP endpoint maps OpenLLMetry and Traceloop attributes, so existing instrumentation can often be repointed. Or add LangSmith's SDK wrappers for OpenAI and Anthropic. If you relied on Helicone's gateway for spend limits, LangSmith's LLM Gateway is also switched on by changing the provider base URL, but it is in beta and does not list caching or failover. Export prompts and datasets before you switch, since nothing migrates them automatically.
As of 20 October 2026, Helicone lists Hobby free with 10K requests and 7-day retention, Pro at $79 a month and Team at $799 a month, but new customers cannot sign up for any tier. LangSmith's Developer tier is free for one seat and 5k base traces a month, Plus is $39 per seat a month with 10k base traces, and the per-trace overage rate is not published. See Helicone's pricing and LangSmith's pricing.
Helicone's free Hobby tier includes 1 GB of storage, one seat and 7-day retention, and Pro includes unlimited seats with one month of retention, but those figures only matter to accounts that already exist. For a new team the useful comparison is LangSmith against another actively developed platform, not against Helicone's price list.
We sell the layer this section describes. Read it with that in mind.
Stated plainly, since Prefactor publishes this page: Prefactor is an agent evaluation platform that scores each step of a run and acts on what it finds through runtime policies on tool calls and data access, approval holds for risky actions, and a kill switch for a run, an agent or a team. Helicone can block prompt injection for OpenAI models and LangSmith's gateway can cap spend and redact model calls, but neither holds a tool action for a person. If you are leaving Helicone, Prefactor accepts OpenTelemetry traces. It has no gateway, no prompt playground or prompt registry, and it is hosted. See the full comparison of 15 platforms.
Reviewed against public product and documentation pages on October 20, 2026. If a vendor has changed a feature, product name, or positioning since then, send a correction and we will update it. Numbered source links in the page body point to the ordered sources below.
Prefactor scores each step of every agent run and can hold or stop an action when a score says it should not happen. Start on the free tier: 5,000 spans a month, no credit card required.
Prefactor helps teams observe, evaluate, and improve their AI agents in production, across every framework and provider.