Helicone Alternatives
Seven tools to replace <em>Helicone</em>, compared on price, gateway features and how deep their tracing goes.
Langfuse is the strongest Helicone alternative in 2026 for most teams: it is open source under MIT for its core, free to self-host, starts at $29 a month on its cloud, publishes a step-by-step guide for teams coming from Helicone, and now has its own AI gateway in public beta. If what you really used Helicone for was the proxy (one base URL, caching, rate limits, fallbacks), LiteLLM is the self-hosted pick and Cloudflare AI Gateway the hosted one, with core features free. Arize Phoenix suits teams standardized on OpenTelemetry, LangSmith suits LangChain and LangGraph shops, Braintrust suits teams whose real problem is evaluation, and Bifrost suits teams that want a fast open source gateway written in Go.
Helicone still works. Its Hobby plan logs 10,000 requests free, Pro is $79 a month and Team $799 a month, and the code stays under Apache 2.0. Teams look elsewhere because Mintlify bought the company in 2026 and put the product in maintenance mode: security updates, bug fixes and new models, but no new features. Its site now says Helicone is part of Experiential Labs, with details to come. Below, each alternative is read the way a buyer reads it: what it costs, whether it sits in the request path, and how much work the move takes.
The ranking is editorial: a sponsored placement is always flagged on its card, and no affiliate link changes a score.
Why teams look past Helicone
If you are new to the category: an LLM observability tool records every call your app makes to a model (the prompt, the answer, tokens, cost, latency, errors) so you can debug, measure and cut spend. Some tools sit in the request path as a proxy or gateway: you change one base URL and every call flows through them. Others read traces sent by an SDK or OpenTelemetry and never touch live traffic. Helicone started as the first kind and became popular because the setup was one line. Here are the reasons we found most often, in the guides and in user feedback, for looking at other tools, each with the fact behind it.
The product is in maintenance mode
Mintlify bought Helicone in 2026 and said the product would keep operating in maintenance mode, with security updates, bug fixes and new models still shipping. Feature work has stopped, and Mintlify itself says it will help every customer migrate to another platform. Helicone's site now adds that it is part of Experiential Labs, with details to come. For a tool that sits between your app and every model call, an unclear owner and a frozen roadmap are reason enough to plan a move, even if nothing is broken today.
Usage pricing on top of the plan
Pro is $79 a month and Team $799 a month, but both say usage-based pricing applies: each paid plan includes 10,000 requests and 1 GB of storage, then you pay per use. The jump from Pro to Team is $720 a month, mostly for 5 organizations, SOC 2 and HIPAA, and three months of retention instead of one. In a late-2024 thread, one team said its bill jumped from nothing to a real monthly line in a single billing cycle once volume tiers kicked in, so model your volume before you upgrade.
Short retention on the lower plans
Hobby keeps data for 7 days, Pro for 1 month and Team for 3 months; only Enterprise keeps it forever. If you want to compare this quarter's cost per user with last quarter's, you need Team at $799 a month or an export routine. Langfuse keeps 90 days on Core at $29 a month and 3 years on Pro at $199 a month.
Request-level logs, thin agent traces
Helicone logs requests, and groups them into sessions with custom headers. That works for a chatbot. For an agent that calls five to ten tools per run, reviews gathered in 2026 describe the request viewer as cluttered once a run reaches five to ten tool calls. Langfuse, Phoenix and LangSmith model a run as a trace with nested spans, which is what you need to see which step failed.
A proxy is one more hop in the request path
In proxy mode, the cloud version routes every call through Helicone's servers, so it adds latency and a point of failure. One guide puts the overhead at 50 to 80 ms per call, and practitioners who measured it in 2024 saw 40 to 120 ms, mostly 60 to 80. For a voice agent or a streaming chat that matters. SDK-based tools such as Langfuse, Phoenix or LangSmith send traces asynchronously and stay out of the path.
Evaluations stay basic
Helicone has scores, datasets, prompts and a playground, but the guides we read agree that its evaluation tooling is thinner than that of evaluation-first platforms. If your next step is running LLM-as-a-judge scorers on production traffic and catching regressions before a release, Braintrust, LangSmith or Langfuse were built for that loop.
Langfuse
Langfuse is an open source LLM engineering platform: tracing of LLM apps and agents with nested observations, session and user tracking, token and cost tracking, prompt management, datasets and evaluations. It is built on OpenTelemetry, has Python and JavaScript SDKs, and accepts proxy-based logging through its own AI gateway or LiteLLM. The core is MIT licensed, except for the enterprise folders, so you can self-host it with Docker Compose, Helm or Terraform. ClickHouse bought the company in 2026 and has kept the license and the cloud running. Langfuse publishes a migration guide for Helicone users: switch tracing, map users, sessions and properties, move prompts, then decide what to do with history. See our list of Langfuse alternatives if you want its own competitors.
How the price reads against Helicone: Hobby is free with 50k units a month, 30 days of data and 2 users. Core is $29 a month with 100k units, then $8 per 100k, 90 days of data and unlimited users. Pro is $199 a month with 3 years of data, and Enterprise $2,499 a month. That is $50 a month below Helicone Pro, with three times the retention. The catch is the unit: a unit is every trace, every observation inside it and every score, so an agent run is several units, not one request (see the traps in the buyer's guide). Its new gateway is still a public beta: it accepts OpenAI and Anthropic API formats, and Amazon Bedrock, Azure, Google AI Studio, OpenRouter and xAI were added as providers in its first weeks.
- MIT core, free to self-host
- Official Helicone migration guide
- Prompts, datasets and evals in the same tool
- ✓Core at $29 a month with unlimited users
- ✓90 days of data on Core, 3 years on Pro
- ✓OpenTelemetry, SDKs and proxy logging all accepted
- ✗Billed per unit, so agent traces count several times
- ✗Its own gateway is still a public beta
- ✗Hobby keeps only 30 days and 2 users
| Criterion | Langfuse | Helicone |
|---|---|---|
| Starting paid plan | $29 a month (Core) | $79 a month (Pro) |
| Retention on that plan | 90 days | 1 month |
| Billing unit | Units (traces, observations, scores) | Requests and storage |
Pick Langfuse if you want one tool for traces, prompts and evals and the option to self-host. Pick LiteLLM or Cloudflare instead if the proxy features were the only reason you used Helicone.
LiteLLM
LiteLLM is an open source AI gateway and Python SDK: one OpenAI-compatible endpoint in front of 140+ providers, virtual keys per team, project or app, budgets and RPM or TPM limits, load balancing and guardrails. It does not try to be a full tracing tool; it logs to Langfuse, Arize Phoenix, LangSmith or any OpenTelemetry backend, which is why the most common replacement for Helicone is LiteLLM plus Langfuse. The code is MIT, except the enterprise directory, and the open source version is self-hosted: you run the server, the database and the upgrades.
How the price reads against Helicone: the open source version is $0 and includes provider integrations, logging, virtual keys, budgets, teams, load balancing and guardrails. Enterprise adds SSO, JWT auth, audit logs, support and custom SLAs, on quote. Your real cost is the server and the engineer who maintains it. One warning: two LiteLLM releases published on PyPI in 2026 were compromised for about 40 minutes before PyPI quarantined them; the official Docker image was not affected. Pin your version and prefer the image.
- 140+ providers behind one OpenAI-compatible API
- Budgets and rate limits per key and team
- Logs to Langfuse, Phoenix, LangSmith and OTel
- ✓$0 for the full gateway feature set
- ✓Keeps the one-base-URL model of Helicone
- ✓Hard budget caps stop requests at the limit
- ✗Self-hosted: you own uptime and upgrades
- ✗Needs a separate tool for rich traces and evals
- ✗SSO and audit logs are Enterprise features
| Criterion | LiteLLM | Helicone |
|---|---|---|
| Price | $0 open source | $79 a month (Pro) plus usage |
| Hosting | Your servers | Cloud or self-hosted |
| Tracing UI | Sends logs to another tool | Built in |
Pick LiteLLM if you want the proxy back under your control and already plan to pair it with a tracing tool. Pick Cloudflare AI Gateway instead if nobody on the team wants to run another server.
Cloudflare AI Gateway
Cloudflare AI Gateway is a hosted proxy on Cloudflare's network: you point your calls at a gateway URL and get dashboard analytics, caching, rate limiting, request retries and model fallback. It also offers DLP scanning, free on all plans with two predefined profiles, guardrails billed as Workers AI inference, Logpush exports, and Unified Billing, which lets you pay for several providers with Cloudflare credits. It is the closest experience to Helicone's original pitch, one line of code, run by a company whose roadmap is not in question.
How the price reads against Helicone: the core features are free on all plans. Log storage is where the limits sit: accounts that created their first gateway before a cut-off in 2026 keep Legacy Logs, with 100,000 logs in total across all gateways on Workers Free and 10,000,000 logs per gateway on Workers Paid; newer accounts store logs in Workers Logs, with its own pricing and retention. Unified Billing adds 5% to credits, so a $100 purchase is charged $105. The limit is depth: it is a gateway with logs and analytics, not a tracing and evaluation platform.
- Analytics, caching and rate limiting free
- Retries and model fallback in the gateway
- Free DLP scanning with two predefined profiles
- ✓$0 platform fee for the core
- ✓No server to run
- ✓Provider prices passed through with no markup
- ✗Workers Free stores 100,000 logs in total on Legacy Logs
- ✗5% fee on Unified Billing credits
- ✗Little in the way of traces, prompts or evals
| Criterion | Cloudflare AI Gateway | Helicone |
|---|---|---|
| Platform fee | Free core features | $79 a month (Pro) |
| Setup | Change the base URL | Change the base URL |
| Evaluations | Not the focus | Scores and datasets |
Pick Cloudflare AI Gateway if you want Helicone's proxy model for free and can live with logs instead of traces. Pick Langfuse instead, or next to it, when you need to debug agents step by step.
Arize Phoenix
Arize Phoenix is an observability and evaluation tool built on OpenTelemetry: it shows each trace as nested spans, with retrieval steps, tool calls and evaluations attached. Several of the guides we read single it out for RAG pipelines. Phoenix is free to self-host under the Elastic License 2.0, which lets you run it but is a source-available license, not an OSI open source one. Arize AX is the managed platform on top, with issue detection (Signal) and enterprise deployment.
How the price reads against Helicone: AX Free gives 25k spans, 1 GB of ingestion and 15 days of retention a month with unlimited users and evals. AX Pro is $50 a month with 50k spans, 10 GB and 30 days. Enterprise is custom, SaaS or self-hosted. That is $29 a month below Helicone Pro, but spans are counted per step: a ten-step agent run is ten spans or more. In 2025, some developers reported spans classified as unknown when sent through plain OTel, so test your instrumentation before you move.
- Native OpenTelemetry and OpenInference
- Strong view of retrieval and tool steps
- Free self-hosting of Phoenix
- ✓AX Pro at $50 a month
- ✓Unlimited users and evals on AX Free
- ✓No proxy in the request path
- ✗Elastic License 2.0, not OSI open source
- ✗Spans add up fast on agent runs
- ✗Some users find the interface heavy
| Criterion | Arize Phoenix | Helicone |
|---|---|---|
| Starting paid plan | $50 a month (AX Pro) | $79 a month (Pro) |
| Ingestion | OpenTelemetry traces | Proxy or async logging |
| License | Elastic License 2.0 | Apache 2.0 |
Pick Phoenix if your stack already speaks OpenTelemetry and RAG quality is your main worry. Pick Langfuse instead if you want an OSI open source license and prompt management in the same place.
LangSmith
LangSmith is LangChain's platform for tracing, evaluating and deploying agents. With LangChain or LangGraph, tracing turns on with environment variables, and every chain, tool call and model call appears as a step in the trace. It also covers evaluations on datasets, and Plus adds access to Agent Server for deploying agents. It works without LangChain through its SDK and OpenTelemetry, but that is where it shines least. Read our LangSmith alternatives page for its own competitors.
How the price reads against Helicone: Developer is free for 1 seat with 5k base traces a month. Plus is $39 per seat a month with 10k base traces included, then pay as you go. Two seats on Plus cost $78 a month, about Helicone Pro, but Helicone Pro has unlimited seats: at five engineers LangSmith is $195 a month. Seats are billed monthly with no credit for removed seats, so add people carefully.
- Zero-code tracing for LangChain and LangGraph
- Evaluations on datasets
- Agent deployment on the same platform
- ✓Free plan with 5k base traces a month
- ✓Deep trace view for chains and agents
- ✓Startup credits of up to $10,000
- ✗Priced per seat, unlike Helicone Pro
- ✗Best value only inside the LangChain stack
- ✗No credit for removed seats
| Criterion | LangSmith | Helicone |
|---|---|---|
| Paid plan | $39 per seat a month | $79 a month, unlimited seats |
| Free volume | 5k base traces a month | 10,000 requests a month |
| Best fit | LangChain and LangGraph | Any provider SDK |
Pick LangSmith if your agents run on LangChain or LangGraph and a small team uses it. Pick Langfuse instead if you are framework-agnostic or have many people who only need to read traces.
Braintrust
Braintrust is an evaluation and observability platform: experiments that compare prompts and models on datasets, scorers (LLM-as-a-judge, autoevals or your own code), a playground, and tracing of production calls that you can turn into new test cases. It is the tool the guides put first when the problem is quality, not cost. Enterprise adds on-prem or hosted deployment for privacy-sensitive data. See our Braintrust alternatives for the other evaluation platforms.
How the price reads against Helicone: Starter is $0 a month with $10 of model credits, 1 GB of processed data then $4 per GB, 10k scores then $2.50 per 1,000, 14 days of retention and unlimited users. Pro is $249 a month with $100 of model credits, 5 GB then $3 per GB, 50k scores then $1.50 per 1,000 and 30 days of retention. That is $170 a month above Helicone Pro, so it pays only if evals are what you buy it for. Qualifying startups can get 6 to 12 months of Pro free.
- Experiments, scorers and playground in one loop
- Production traces become test cases
- Unlimited users on every plan
- ✓Free Starter with 10k scores a month
- ✓Unlimited users, projects and datasets
- ✓6 to 12 months of Pro free for qualifying startups
- ✗Pro at $249 a month is a big step
- ✗14-day retention on Starter
- ✗No gateway to replace the Helicone proxy
| Criterion | Braintrust | Helicone |
|---|---|---|
| Paid plan | $249 a month (Pro) | $79 a month (Pro) |
| Core strength | Evaluations | Request logging and cost |
| Users | Unlimited | Unlimited on Pro |
Pick Braintrust if you are replacing Helicone because evaluations became the bottleneck. Pick Langfuse instead if you want evals and tracing at a lower entry price.
Bifrost
Bifrost is an open source AI gateway written in Go by Maxim AI: one API in front of many providers, spending limits across teams, projects and models, a central place to govern MCP tool connections, OpenTelemetry export and a built-in dashboard. Maxim says it adds 11 microseconds of overhead at 5,000 requests per second on a t3.xlarge machine (its headline figure is 20) and calls it 50 times faster than LiteLLM on P99 latency; both are the vendor's own benchmarks. It ships as a single process you can start with npx or Docker, and offers a compatibility path for teams coming from LiteLLM.
How the price reads against Helicone: Bifrost is free under Apache 2.0, the same license as Helicone's code. Enterprise features and support are on quote, with a 14-day free trial; Maxim does not show an amount for them. Like LiteLLM, the real cost is hosting and upkeep. It is a young project, so check its issue tracker and release pace before you put it in front of production traffic.
- Go binary built for low overhead
- Budgets per team, project and model
- MCP connections governed in the gateway
- ✓$0 under Apache 2.0
- ✓Quick start with npx or Docker
- ✓OpenTelemetry export to your tracing tool
- ✗Speed figures are the vendor's own benchmarks
- ✗Self-hosted: you own uptime
- ✗Enterprise price not published
| Criterion | Bifrost | Helicone |
|---|---|---|
| Price | $0 open source | $79 a month (Pro) plus usage |
| License | Apache 2.0 | Apache 2.0 |
| Focus | Gateway and governance | Gateway and observability |
Pick Bifrost if gateway speed and MCP governance matter more than a mature ecosystem. Pick LiteLLM instead if you want the widest provider list and the larger community.
7 Helicone alternatives compared
Scores are our editorial score out of 5, the plain average of the five sub-scores on each card. The last column answers the first question to ask when you replace Helicone: does the tool sit in your request path as a gateway, or does it read traces from an SDK as a tracing platform? Helicone did both, so most teams end up with one of each.
| Best for | Gateway or tracing | Free plan | Team size | Visit | ||||
|---|---|---|---|---|---|---|---|---|
| 1 | Langfuse | Best overall Helicone alternative | Tracing platform; gateway in beta | 4.5/5 | Hobby free (50k units a month); Core $29 a month; Pro $199 a month; self-hosting free | ✓ | Startups to enterprise | Visit → |
| 2 | LiteLLM | Best self-hosted gateway to replace the Helicone proxy | Self-hosted gateway | 4.4/5 | Open source $0, self-hosted; Enterprise on quote (SSO, audit logs, support) | ✓ | Platform teams | Visit → |
| 3 | Cloudflare AI Gateway | Best hosted proxy with a free core | Hosted gateway | 4.3/5 | Core features free on all plans; 5% fee on Unified Billing credits; log storage depends on your Workers plan | ✓ | Startups and SMB | Visit → |
| 4 | Arize Phoenix | Best for OpenTelemetry-first teams | OTel tracing platform | 4.2/5 | Phoenix free to self-host; AX Free 25k spans a month; AX Pro $50 a month | ✓ | Engineering teams | Visit → |
| 5 | LangSmith | Best for LangChain and LangGraph teams | SDK tracing platform | 4.2/5 | Developer $0 (1 seat, 5k base traces a month); Plus $39 per seat a month (10k traces) | ✓ | LangChain teams | Visit → |
| 6 | Braintrust | Best for evaluation-first teams | Evaluation platform | 4.1/5 | Starter $0 a month (1 GB, 10k scores); Pro $249 a month (5 GB, 50k scores) | ✓ | AI product teams | Visit → |
| 7 | Bifrost | Best open source gateway for raw speed | Self-hosted gateway | 4.0/5 | Free and open source (Apache 2.0); Enterprise on quote, with a 14-day free trial | ✓ | High-volume teams | Visit → |
Prices are list prices in US dollars for 2026, monthly unless stated. Usage above the included volume is billed on top on Langfuse, Arize, LangSmith and Braintrust.
Is your tool missing from this page? Email us: hello@hackceleration.com
You're leaving Helicone because... take this
Tracing, prompts and evals, MIT core.
You only used the proxy and want to self-host itLiteLLMFree gateway with keys and budgets.
You want a hosted proxy and no monthly feeCloudflare AI GatewayAnalytics, caching and rate limits free.
Your stack already emits OpenTelemetryArize PhoenixOTel-native traces and evals.
Your agents run on LangChain or LangGraphLangSmithTracing with environment variables.
Evaluations are your real bottleneckBraintrustExperiments, scorers, production cases.
You need gateway speed and MCP governanceBifrostGo gateway under Apache 2.0.
How to choose a Helicone alternative
LLM observability works in three layers. First, capture: either a gateway in the request path (change the base URL, every call is logged) or an SDK or OpenTelemetry instrumentation in your code that sends traces in the background. Second, storage and views: each call becomes a request (Helicone, Cloudflare) or a trace made of nested spans or observations (Langfuse, Phoenix, LangSmith, Braintrust), with cost, latency, user and session attached. Third, action: alerts, prompt versions, datasets and evaluations that turn what you saw into a fix. Helicone covered the first layer very well and the third only lightly. Decide first whether you are replacing the gateway, the tracing, or both. For the wider market, see our rankings of the best LLM observability tools and the best AI gateways.
Before you compare quotes, write down four numbers: requests per month, average steps per agent run, how many people need access, and how many days of history you must keep. Then ask each vendor for its billing unit (requests, units, spans, traces, seats or GB), its retention on the plan you would buy, whether it can import or accept Helicone-style headers, and whether you can export raw data without a ticket. Other names come up in the guides. Portkey was a strong hosted gateway, but Palo Alto Networks completed its purchase in 2026 and portkey.ai/pricing now sends you to a contact form for Prisma AIRS AI Gateway, with no public plan; see our Portkey alternatives. OpenRouter is a model marketplace with a 5.5% fee on credits on its Standard plan (8% on Business), see our OpenRouter alternatives. W&B Weave, Datadog LLM Observability, TrueFoundry, Kong AI Gateway and PostHog LLM analytics are also cited; if you already pay for PostHog or Datadog, start there.
Gateway, tracing, or both
If your code calls OpenAI or Anthropic directly and you only want logs and costs, a gateway is enough: LiteLLM, Cloudflare or Bifrost. If you run agents or RAG, you need traces: Langfuse, Phoenix, LangSmith or Braintrust. Many teams pair LiteLLM for routing and budgets with Langfuse for traces, since LiteLLM sends its logs straight to Langfuse.
Trap 1: requests are not units
Helicone counts requests: its free plan and each paid plan include 10,000 a month. Langfuse counts units: the trace, every observation inside it and every score. An agent run with one trace, nine steps and one score is 11 units, so the 50k units of the Hobby plan cover about 4,545 such runs, and the 100k of Core about 9,090, before the $8 per 100k overage. Count your steps per run before you size a plan.
Trap 2: the hidden 5% and the 100,000-log cap
Cloudflare AI Gateway is free, but Unified Billing adds 5% to every credit purchase: $1,000 of model credits costs $1,050, so $600 a year at that monthly spend. Pay providers directly with your own keys if you want zero markup. And on Legacy Logs, Workers Free stores 100,000 logs in total across all gateways, not per month: a team at 10,000 requests a day fills it in 10 days.
Trap 3: self-hosting Helicone while you decide
Self-hosting Helicone is a fair bridge, but read its Docker guide. The all-in-one container bundles PostgreSQL, ClickHouse and MinIO, and a container restart wipes all data unless you mount volumes. The image no longer proxies LLM calls on its own: you deploy Helicone's AI Gateway next to it. Port 8585 needs no authentication for proxied requests, so it must sit behind firewall rules. The default auth secret is change-me-in-production, and the guide tells you to change it.
Export before you move
Helicone's export CLI (npx @helicone/export) pulls requests by date range and custom property, with full bodies and JSON, JSONL or CSV output, and a separate EU region. Without a start date it exports only the last 30 days, so set one. Langfuse's advice is to export early, archive the file, and reingest only what you need: reingested traces are billed as new units, and Langfuse does not deduplicate IDs sent twice, so checkpoint the import. Most teams keep history as an archive and start fresh with live traffic.
Run both for a day
Swap the base URL or add the SDK on one service, run a day of real traffic through the old and the new tool in parallel, and reconcile token counts and costs with your provider invoice. Remap every Helicone header (Helicone-User-Id, Helicone-Session-Id, Helicone-Property-[Name]) to the new tool's fields, or your user and session views will be empty.
- Are we replacing the gateway, the tracing, or both?
- What is the billing unit: requests, units, spans, traces, seats or GB?
- How many steps does one agent run produce?
- How many days of history must we keep?
- Can we self-host, and under which license?
- Can we export raw data without a support ticket?
Helicone alternatives, your questions
Is Helicone still safe to use after the Mintlify acquisition?
Yes for now, with a deadline in mind. Mintlify said Helicone keeps operating in maintenance mode, with security updates, bug fixes and new models, but no new features. The real risk is your history: request logs, prompts and configuration can only be exported while the service stays up, so run the export now even if you migrate later.Who owns Helicone now?
Mintlify bought Helicone in 2026 and its team joined Mintlify in San Francisco. Since then, Helicone's home and pricing pages carry a banner saying Helicone is now part of Experiential Labs, with details to come soon. Until those details are public, treat ownership as unsettled in your vendor review.Is Helicone open source?
Yes. The Helicone repository is under the Apache 2.0 license and has about 6,300 stars on GitHub, so you can self-host it with Docker even if the hosted service changes. Helm charts for Kubernetes sit in a separate public repository, licensed Apache 2.0 with the Commons Clause, while the main README still tells enterprise workloads to ask enterprise@helicone.ai for a production-ready chart.What are the limits of Helicone's free plan?
Beyond 10,000 requests a month, the main one is speed: Hobby ingests 10 logs per minute, against 1,000 on Pro and 15,000 on Team. It also has 1 seat, 1 organization, 1 GB of storage and 7 days of retention, so it suits a prototype, not a production app with bursts of traffic.Does Helicone add latency to my LLM calls?
On the cloud proxy, yes, a little. One guide estimates 50 to 80 ms per call, and practitioners measured 40 to 120 ms in 2024. Helicone itself offers an async integration that logs calls through OpenLLMetry without placing Helicone in your app's critical path, so an issue on its side cannot take your app down.Can I keep Helicone for live traffic and add another tool for evals?
Yes, and some teams do it as a transition. The usual pattern is Helicone for request and cost logs and a tracing tool for agent debugging and evaluations. The risk is paying twice and splitting data across two tools, so set a date to retire one of them.What does a Helicone migration involve?
Usually one code change and one data decision. You swap the base URL or add an SDK, remap user, session and property headers, recreate prompt versions in the new tool, and decide whether to reingest old requests. Langfuse's guide suggests keeping the old gateway during a comparison window.Does Langfuse have a gateway like Helicone's proxy?
Yes, since late 2026, in public beta. The Langfuse AI Gateway forwards requests in the OpenAI or Anthropic API format with provider keys your organization stores in Langfuse, hands out gateway keys per person or service, and caps spend with budgets per key, user or organization. It launched with OpenAI and Anthropic and has since added Amazon Bedrock, Azure, Google AI Studio, OpenRouter and xAI; each generation it records counts as a billable unit.Is Arize Phoenix open source?
Phoenix is free to self-host, but its license is the Elastic License 2.0, a source-available license rather than an OSI-approved open source one. It lets you run and modify Phoenix, but not offer it to others as a managed service. If your policy requires OSI open source, Langfuse (MIT core) and Helicone (Apache 2.0) qualify.