When LangSmith or Helicone fits a LangSmith replacement
Helicone fits teams that want request logging, cost and latency tracking, sessions and routing in one self-hosted component, and that can accept a smaller evaluation scope. LangSmith fits teams where evaluation depth matters most and where a LangChain-centered workflow is already in place. The renewal, licensing change or residency requirement that brought you here does not decide the question on its own, because the two products solve different parts of the same problem. Start by listing which of your workloads depend on datasets, experiments, online evaluators and annotation queues, since those are where the gap is widest.
Three factors should drive the decision. The first is evaluation needs: if release gates depend on LangSmith-style experiments and human review queues, plan to rebuild that layer with another tool, such as promptfoo in CI, whatever gateway you pick. The second is LangChain dependence: if your applications are built on LangChain or LangGraph, check how much of your tracing instrumentation depends on LangSmith callbacks, and whether the Helicone base URL or OpenLLMetry approach fits your code paths. The third is tolerance for project risk. Mintlify announced on March 3, 2026 that it is acquiring Helicone, and both companies say Helicone will continue in maintenance mode, with security updates, bug fixes and new models still shipping. Ask Mintlify in writing what the roadmap, support terms and migration help look like, and confirm which Helicone release you will pin, since the latest tagged release is v2025.08.21-1 from August 21, 2025.
Before you commit, run a short proof of concept in your own network. Deploy the Docker Compose setup, route one non-production application through the gateway, and confirm that the Next.js frontend, the Cloudflare Workers logging proxy, Jawn, Supabase, ClickHouse and MinIO all meet your hosting, backup and access-control standards. Then compare the result against your current LangSmith workflow for the evaluation tasks you listed. If Helicone covers logging and routing but leaves a gap in evaluation, the usual answer is a split design rather than forcing one tool to do everything. Document that decision and the ownership of each component so auditors and platform teams can follow it.
Which option fits which buyer situation
This table maps common buyer situations to the better fit between LangSmith and Helicone. Read each row as a starting point, then check the named capability against the current release and your own environment, because Helicone's roadmap has changed since its acquisition announcement.
| Buyer situation | Better fit | Why |
|---|---|---|
| Gateway and tracing in one self-hosted component | Helicone | Its gateway routes requests to 100+ models through the OpenAI API and logs them through the same path, with cost, latency and session tracking included. |
| Deep datasets, online evaluators and annotation workflows | LangSmith | Evaluation depth is the main reason to choose LangSmith. Confirm the specific evaluator and experiment features you need against its current docs. |
| Heavy LangChain and LangGraph use | LangSmith | It is built by the same company as those frameworks, so callbacks and trace views match them closely. Helicone can also log through OpenLLMetry, which needs more setup. |
| Self-hosting without an enterprise contract | Helicone | Helicone is Apache-2.0 licensed and ships a Docker Compose setup. Self-hosted LangSmith requires the Enterprise add-on and a license key. |
| Long-term supported roadmap | Neither without confirmation | Mintlify announced its acquisition of Helicone in March 2026 and says Helicone moves to maintenance mode, with security fixes and new models still shipping. Ask Mintlify for the current support commitment before you standardize. |
| Stack has a cloud region requirement | LangSmith | LangSmith Cloud offers regions in the US, EU, and APAC. Helicone's hosting location depends entirely on the deployment you run. |
| Small team with one gateway and basic cost visibility | Helicone | Fewer components to run than a full evaluation stack, and the gateway covers routing and fallbacks from day one. |
How Helicone captures requests and where the gateway sits
Helicone offers two ways to capture requests, and the choice decides whether a component sits in your request path. In the gateway path, the application changes its client base URL to the Helicone gateway, so every model call passes through infrastructure your team runs. In the asynchronous path, the application keeps calling the model provider directly and instruments itself with OpenLLMetry, which sends spans to Helicone after the fact. The gateway path also gives routing and automatic fallbacks, which the asynchronous path does not. Before you choose, confirm which of these your deployed version supports end to end, since the gateway is the feature that carries the most operational weight.
The trust boundary differs sharply between the two paths. With the gateway, prompts, completions and provider API keys all transit a proxy you must secure, monitor and keep available, so an outage or misconfiguration there can break production traffic, not just dashboards. With OpenLLMetry, a logging failure should not block a model call, but prompt and response content still leaves the application for the Helicone collector, so the data-residency question remains. In a self-hosted Helicone deployment, the logging proxy is a Cloudflare Workers component by default, which your security team should review against your own edge and network policies before you place it in front of regulated workloads.
LangSmith takes a different approach. Its tracing is largely driven by the LangChain callback model, where instrumentation lives in the application code and traces are sent from inside the process, and a LangChain or LangGraph app gets much of this with little extra wiring. Helicone's model is closer to a network boundary, which suits teams running mixed SDKs or several frameworks, but it asks more of the platform team to operate. For a platform team, the practical check is to map every service that calls a model, decide which can tolerate a gateway hop, and confirm how each service's keys and content will be handled in your environment.
Data location, hosting models and compliance implications
Both products can keep traces in infrastructure you control, but they get there differently. LangSmith's self-hosted path runs through an Enterprise contract, while Helicone's self-hosted stack is open source with several components your team must patch and secure.
LangSmith Cloud regional placement
LangSmith Cloud offers GCP US, GCP EU, GCP APAC and AWS US regions, and data stays in the region where the account signed up. Billing information stays in the US, even for EU accounts. If your residency rule covers billing metadata as well as traces, record that gap before you sign and ask the vendor to confirm how account identifiers are handled.
Self-hosted LangSmith requires a license
Self-hosted LangSmith is an add-on to LangChain's Enterprise plan and needs a license key. It deploys to your Kubernetes cluster with Helm charts, with Terraform also documented. Your team owns the cluster, storage, upgrades and backups, and the license is the commercial gate you must clear before any deployment work begins.
Helicone self-hosting with Docker Compose
Helicone self-hosts with a Docker Compose setup, and its README says a production Helm chart for enterprise workloads is available by contacting Helicone. Confirm whether that chart's terms, support and update cadence fit your environment before choosing it over a chart you build and maintain internally. Compose is suited to evaluation rather than a production footprint without further hardening.
Helicone stack as a patching surface
A self-hosted Helicone deployment runs a Next.js web frontend, a Cloudflare Workers proxy for logging, the Jawn log collection server, Supabase for the application database and auth, ClickHouse for analytics and MinIO for log object storage. Each component holds request data or credentials, so each needs its own patch owner, network policy and backup plan. The Cloudflare Workers proxy also means logs may transit a Cloudflare component, which must be checked against your residency rule.
Helicone maintenance status
Mintlify announced on March 3, 2026 that it is acquiring Helicone. Both companies say Helicone continues in maintenance mode, with security updates, bug fixes and new models still shipping, and Mintlify offers help moving customers to another platform. Ask in writing what support commitment applies to your deployment and who you contact for a security advisory after the acquisition closes.
Compliance evidence and vendor statements
DeployOpen prepares systems, controls and evidence for compliance assessments, but it does not issue audit reports, attestations or certifications. Independent auditors and certification bodies do that work. Request each vendor's current compliance statements directly, and confirm which controls apply to your self-hosted deployment versus their hosted service before you cite either in an assessment.
Cost components on each side and what drives them
Comparing LangSmith with Helicone means separating software licensing from the infrastructure you run yourself. Each side has different recurring and one-time cost lines, and the largest variable is usually how long you keep logs. The cards below list each component and what to request in a written quote.
LangSmith license and Enterprise add-on
Self-hosted LangSmith is an add-on to LangChain's Enterprise plan and needs a license key, so the subscription is the first cost line to confirm. Ask whether the quote covers the license for the Helm or Terraform deployment, the number of seats, and whether trace volume or retention is billed separately. Treat the license as recurring, and ask how renewals and price changes are handled over a multi-year term.
Helicone software and support
Helicone's repository is Apache-2.0, so the software itself carries no license fee. The README says a production Helm chart for enterprise workloads is available by contacting Helicone, which means that chart and any support terms are a negotiated item. Ask whether the chart, upgrade help and support are included, and what the written scope says about maintenance mode.
Infrastructure for each stack
A self-hosted Helicone deployment runs a Next.js frontend, a Cloudflare Workers proxy, the Jawn log server, Supabase, ClickHouse and MinIO, so the hosting footprint is larger than the component count suggests. LangSmith's self-hosted deployment runs on your Kubernetes cluster. Compare the compute, managed database and object storage each side needs, and whether Cloudflare Workers usage is part of your budget.
Storage growth with log retention
Log volume, prompt payloads and retention period drive storage on ClickHouse and MinIO, and that cost recurs for as long as you keep data. Ask the vendor or partner how retention is configured, how old logs are pruned or archived, and what sizing assumptions sit behind the quote. Retention is usually the largest variable, so model it against your real request volume before comparing totals.
Upgrades, patching and maintenance
Upgrades for a self-hosted stack are recurring labor on both sides, covering Helm or Compose changes, database migrations and security patches across several services. Helicone's maintenance mode means security updates, bug fixes and new models still ship, but feature work does not. Ask who applies patches, how often, and whether a retainer covers upgrades and testing in the scope.
One-time migration and setup
Initial deployment, gateway base URL changes or OpenLLMetry instrumentation, and any data movement are one-time costs. They are usually larger than the first year's software line when several applications depend on the current tracing setup. Ask for a fixed-scope estimate listing each application to reroute, the data to move, and the testing needed before cutover.
What running each platform takes day to day
Helicone is a multi-service stack you would operate yourself. Its self-hosted deployment runs a Next.js web frontend, a Cloudflare Workers proxy for logging, the Jawn log collection server, Supabase for the application database and auth, ClickHouse for analytics and MinIO for log object storage. That is at least three stateful components to back up, patch and size (Supabase, ClickHouse and MinIO), plus the proxy path that every logged request crosses. The Docker Compose setup is the documented starting point, and the README says a production Helm chart for enterprise workloads is available by contacting Helicone. Before committing, ask the vendor what that chart covers, how it handles ClickHouse and Supabase upgrades, and who supports it when a request fails at the proxy.
The upgrade path needs a decision before you build on Helicone. The latest tagged GitHub release is v2025.08.21-1, published on August 21, 2025, and Mintlify announced on March 3, 2026 that it is acquiring Helicone. Both companies say Helicone moves into maintenance mode, with security updates, bug fixes and new models still shipping. Maintenance mode is a reasonable position for a tool you run, but it means you should track the repository's commits and release notes yourself, test each dependency bump in staging, and confirm how security fixes reach self-hosted installs. The Mintlify-hosted roadmap and support terms are the points to confirm directly with Mintlify and Helicone, in writing, rather than assuming they match the README.
LangSmith moves the work to your platform team in a different shape. Self-hosted LangSmith deploys to your Kubernetes cluster with Helm charts, with Terraform also documented, and requires an Enterprise license key. Your platform team owns the cluster, ingress, storage sizing, upgrades to chart versions and the license lifecycle, while LangChain supplies the software and the entitlement. The load is fewer custom components to assemble but more dependence on the Kubernetes operating model your team already runs. Ask LangChain which chart versions are supported for your cluster version, how upgrades are sequenced and what the license check requires from your network.
How a move from LangSmith to Helicone involves the work
The sequence below takes a team from an inventory of LangSmith usage to a self-hosted Helicone deployment that logs, tracks cost, and routes requests, with a rollback path kept open until the new stack has proven itself.
Inventory LangSmith usage and confirm data export access
Engineering lists every application that sends traces to LangSmith, including LangChain and LangGraph callbacks, datasets, prompts, evaluators, and dashboards that people rely on. Platform or procurement then confirms which LangSmith plan the account is on and whether bulk export to an S3-compatible bucket in Parquet format is available, because the August 3, 2026 change limits it to the Enterprise plan for newer customers and keeps earlier Plus access until February 1, 2027. Done when the inventory is signed off and the export answer is written down, since a missing export path changes how history is kept.
Decide whether the gateway sits in the request path
Architecture and security decide whether Helicone's gateway will route live model traffic or only record it. Routing through the gateway gives one place for model access, fallbacks, and cost controls, but every request then depends on that component's availability and on how it is patched. Logging through asynchronous OpenLLMetry keeps model calls on the existing path. The decision is recorded with the owners of uptime and incident response, and a reviewer confirms that the chosen path matches the data handling rules for prompts and completions.
Deploy the self-hosted Helicone stack in your environment
Platform engineers deploy Helicone with its Docker Compose setup for a test environment, or with the production Helm chart that the README says is available by contacting Helicone. The stack includes a Next.js frontend, a Cloudflare Workers proxy for logging, the Jawn log collection server, Supabase for the application database and auth, ClickHouse for analytics, and MinIO for log object storage. Operators confirm which of these components run inside the client's own cloud account or data center, and they check the Cloudflare Workers piece against the network and residency rules before anything goes live.
Switch logging to base URL or OpenLLMetry
Application teams change the client base URL to the Helicone gateway for routed traffic, or add asynchronous OpenLLMetry instrumentation where requests should stay on their current path. Each service is changed in a branch, and tests confirm that requests succeed, responses are unchanged, and each call appears in Helicone with the expected model, user, and session metadata. Done when a sample of production-like calls from every service shows up in the new dashboards, with no change in application error rates during the test window.
Map sessions, prompts and evaluators to Helicone
Teams translate LangSmith projects and sessions into Helicone sessions and identify the prompts that were versioned in LangSmith's prompt hub. Helicone's prompt playground and versioning are deployed through the gateway, so each prompt version is recreated and linked to the requests that use it. Evaluators need a separate decision: Helicone's features are centered on logging, cost, latency, and prompt versions, so LangSmith datasets, online evaluators, and annotation queues either move to another tool or are rebuilt from exported Parquet data. The output is a mapping sheet that names every prompt, session, and evaluator and its new home.
Validate cost and latency tracking against known values
Finance and engineering compare Helicone's cost and latency numbers with provider invoices and with LangSmith figures for the same period. Discrepancies are traced to model price tables, token counting, cached or streamed responses, and retries. Helicone is in maintenance mode after Mintlify's acquisition announced on March 3, 2026, so the team also confirms that the model price data and security patches it depends on are being kept current under a written support or upgrade plan. Done when cost deltas are explained and the owner of price table updates is named.
Cut over with a documented rollback path
Operations moves production traffic service by service, keeping the LangSmith configuration and the Helicone base URL switch in version control so each change can be reverted in minutes. Before cutover, the team agrees the rollback trigger, such as error rate, missing traces, or cost variance, and who makes the call. LangSmith remains read-only for the retention window the organization needs, and exported Parquet files are stored in the organization's bucket with access logs. Done when every service runs on Helicone, the rollback has been rehearsed in a test environment, and the decommission date for LangSmith is approved.
Questions buyers ask before choosing Helicone or LangSmith
Is Helicone still maintained after the Mintlify acquisition?
Helicone is still maintained, but in a narrower form. Mintlify announced on March 3, 2026 that it is acquiring Helicone, and both companies say Helicone will move to maintenance mode. Security updates, bug fixes and new model support will keep shipping, and Mintlify is offering to help customers move to another platform. Feature development is not promised. The repository is licensed under Apache-2.0, so a self-hosted copy stays usable and your team can maintain a fork if needed. Before you commit, read Mintlify's announcement and ask both companies for written support terms, a security patch commitment and the model list they will keep current. Also check the Helicone releases page for anything newer than the latest tagged release, v2025.08.21-1, published on August 21, 2025, before you pin a version.
What does self-hosting Helicone require from my team?
Your team needs to operate several components, not one container. Docker Compose is the documented starting point for testing, and the README says a production Helm chart for enterprise workloads is available by contacting Helicone. A self-hosted deployment runs:
- A Next.js web frontend
- A Cloudflare Workers proxy for logging
- The Jawn log collection server
- Supabase for the application database and auth
- ClickHouse for analytics
- MinIO for log object storage
Each piece needs backups, upgrades, monitoring and access control. The Cloudflare Workers proxy is the one to check first, because it sits in the request path and depends on Cloudflare. Confirm whether your network and data policy allow that dependency, or whether an in-cluster alternative is acceptable to your security team.
Can Helicone match LangSmith's evaluation depth?
Probably not out of the box, so treat evaluation as a gap to test. The Helicone material we reviewed centers on request logging, traces and sessions, cost and latency tracking, a prompt playground, prompt versioning and gateway routing with fallbacks. LangSmith's evaluation workflows include datasets, experiments, online evaluators and annotation queues. If your team relies on those, list each workflow and confirm how Helicone handles it in a proof of concept. Where a workflow is missing, plan to rebuild it on a separate tool, such as Langfuse or Opik for evaluations, or promptfoo for checks that gate releases in CI. That split adds a second component to run, so count it in the operating cost.
What drives the cost of Helicone compared with LangSmith?
The cost structures differ in kind. With LangSmith, you pay for the plan or, for self-hosting, an Enterprise add-on that requires a license key from LangChain. With Helicone, the Apache-2.0 license carries no software fee, so the recurring costs are infrastructure and people. Infrastructure covers the Kubernetes or Compose hosts, ClickHouse and Supabase capacity, MinIO storage, and any Cloudflare Workers usage. People cover operations, upgrades and on-call. The main drivers are log volume, retention period, number of environments and high availability requirements. When you ask for a quote, request each cost line separately, including who pays for the Helm chart support, and ask which items recur monthly and which are one-time setup.
Where does my data stay with Helicone compared with LangSmith Cloud?
With a self-hosted Helicone deployment, traces and logs stay wherever you deploy the stack, provided the components are in your environment. LangSmith Cloud runs in GCP US (Iowa), GCP EU (Netherlands), GCP APAC (Sydney) and AWS US (Ohio), and data stays in the region where the account signed up. Billing information stays in the US. For Helicone, confirm the location of the Cloudflare Workers proxy, since requests pass through it before logging, and confirm where any Supabase and MinIO data is stored. Ask your security and compliance teams to map each component to a data category, so the residency answer is written down and can be shown to auditors.
How long does a migration from LangSmith to Helicone take?
No duration is established for this migration, and the time depends on several factors you can measure. The first is instrumentation: applications that change their base URL to the Helicone gateway are simpler than those that need OpenLLMetry asynchronous logging wired into existing code. The second is the number of prompts, versions and sessions to recreate. The third is historical data. LangSmith's bulk export writes trace data to an S3-compatible bucket in Parquet format, but for customers who signed up after August 3, 2026, it is only available on the Enterprise plan. Earlier Plus customers keep access until February 1, 2027. Confirm your plan and export window early, then ask for a written plan with milestones for each environment.
Can our team do the migration without outside help?
Yes, if you already run Kubernetes, Helm and stateful data services in production and have engineers who can own the gateway change. The work splits into four parts: deploying the stack, changing client instrumentation, exporting and mapping historical data, and rebuilding evaluations. Each part needs a named owner and a rollback path. Outside help is useful when the team lacks capacity or wants a second review of security controls. DeployOpen plans, implements, integrates, hardens, backs up, upgrades, documents and supports self-hosted open-source platforms in your own infrastructure. Most engagements are fixed-scope, with scope, deliverables and timelines agreed in writing before work starts, and ongoing support is available as a retainer.
Deciding when evaluation depth outweighs a single self-hosted component
The deciding question is where the value sits. If your LLM work depends on evaluation datasets, online evaluators and annotation queues, LangSmith's depth is hard to match with a logging-first tool, and an open-source platform built around evaluation is a closer fit. Langfuse covers LLM-as-a-judge, annotation queues and dataset experiments, and Opik is Apache-2.0 licensed with its evaluation metrics and prompt library included in the self-hostable platform. Before choosing either, list the evaluators your team runs today and confirm which ones each platform supports natively and which would have to be rebuilt in code.
If the priority is routing and request logging in one self-hosted component, Helicone is a reasonable candidate, but weigh its status first. Its repository is Apache-2.0 licensed, and Mintlify announced its acquisition of Helicone on March 3, 2026, with Helicone moving to maintenance mode. Security updates and bug fixes continue, and Mintlify has offered to help customers move elsewhere. Ask whether your security team accepts a component in maintenance mode, and plan a migration path in case the roadmap changes. Helicone's self-hosted stack is also heavier than its single-tool appearance suggests, since it runs Supabase, ClickHouse and MinIO alongside its gateway.
If you need a gateway in front of several providers, LiteLLM and Portkey are the usual alternatives. The LiteLLM AI Gateway is a self-hosted server with virtual keys, budgets, fallbacks and logging to Langfuse, while LiteLLM is MIT licensed except for its enterprise directory, which carries SSO, SAML and audit logs under a separate license. Portkey's AI Gateway is MIT licensed, but prompt management, advanced logging and PII redaction sit outside the open-source gateway as enterprise features. Pair whichever gateway you pick with an evaluation platform, and confirm each license terms in writing before you standardize.
How DeployOpen helps evaluate, implement or migrate
DeployOpen plans, implements, hardens, documents and supports self-hosted open-source platforms inside your own infrastructure. For a LangSmith vs Helicone decision, we start by writing down which traces, costs, sessions, prompts and evaluations your teams actually depend on, then deploy the candidate platform in your cloud account, data center or air-gapped network.
How we scope
- Inventory which LangSmith traces, prompts, datasets and evaluators must carry over, and which Helicone features cover them
- Confirm whether the gateway must sit in the request path or logging through OpenLLMetry is enough
- Agree the deployment target, data region and retention rules in writing before any build work starts
How we build
- Deploy Helicone with Docker Compose or the enterprise Helm chart, with its Supabase, ClickHouse and MinIO components sized to your log volume
- Test base URL switching and OpenLLMetry logging against representative agent sessions and cost tracking
- Run a parallel evaluation of LangSmith evaluation features that Helicone does not cover, so gaps are measured rather than assumed
How we hand over
- Document the architecture, the gateway configuration, data flows and the backup and restore procedure for your team
- Deliver a written gap register listing evaluation, dataset and annotation capabilities that still need another tool
- Provide runbooks for upgrades, secret rotation and access control, and confirm each with your engineers before sign-off
How we support
- Offer ongoing support, upgrades and continuous testing as a retainer after handover
- Track Helicone's maintenance status and plan any move if the platform's roadmap changes after the Mintlify acquisition
- Scope any follow-on work, such as adding an evaluation platform, as a separate fixed-scope engagement with a written quote
