Who Bought Your LLM Observability Tool? The 2025-2026 Consolidation, and How to Stay Portable

TL;DR
Langfuse is part of ClickHouse, Promptfoo is part of OpenAI, Galileo is part of Cisco, Traceloop is joining ServiceNow, Helicone joined Mintlify and Humanloop shut down. Who owns what in October 2026, the dates the vendors state, and a portability checklist built on OTLP ingest, licenses and self-hosting.
Last updated: October 18, 2026
Short answer: seven LLM tracing and eval tools have changed hands, or say on their own pages that they are changing hands, and the dated events run from May 2025 to May 2026. Langfuse is part of ClickHouse. Promptfoo is part of OpenAI. Galileo is part of Cisco and its pricing now lives on a Splunk page. Traceloop is joining ServiceNow. Weights & Biases Weave is documented and priced under CoreWeave. Helicone joined Mintlify and runs in maintenance mode. Humanloop's team is joining Anthropic and its platform was sunset on September 8, 2025. None of this means you must migrate tomorrow. It does mean the tool you picked is now one product line inside a much bigger company, so the part of your stack that should not belong to anyone is the instrumentation and the eval suite.
Everything below comes from the vendors' own pages: the LLM observability directory we verified on October 18, 2026, where every value carries a verbatim quote, plus each acquirer's or vendor's own announcement for the dates. Where a page gives no date, this post says "no date given".
Who changed hands#
| Tool | Now part of | What the vendor's page says | Date, from a primary source | License and self-host |
|---|---|---|---|---|
| Langfuse | ClickHouse | "since January 2026 we're part of ClickHouse" (README) | Announced Jan 16, 2026 (Langfuse announcement) | MIT Expat outside the ee/ directories; core features free to self-host |
| Promptfoo | OpenAI | "Promptfoo is now part of OpenAI." (README) | Announced March 9, 2026 (OpenAI) | MIT; run locally or self-host |
| Galileo | Cisco (Splunk) | "Galileo is now part of Cisco" (galileo.ai) | Completed May 22, 2026 (Cisco) | License not stated (commercial); the Splunk page lists cloud and self-hosted deployments |
| Traceloop / OpenLLMetry | ServiceNow | "Traceloop is joining Service now!" (pricing banner) | Traceloop post dated March 2026; ServiceNow called it completed on May 5, 2026 (ServiceNow) | OpenLLMetry is Apache 2.0; Traceloop lists on-prem deployment |
| W&B Weave | CoreWeave | Pricing on the "CoreWeave Forge Pricing" page; docs published as CoreWeave Docs | CoreWeave completed the W&B acquisition May 5, 2025 (CoreWeave) | Weave repository is Apache 2.0; self-host not stated |
| Helicone | Mintlify | "Helicone Joins Mintlify" (pricing banner) | Announced Mar 3, 2026 (Helicone) | Apache 2.0; self-hostable |
| Humanloop | Team joining Anthropic; platform shut down | "the Humanloop team is joining Anthropic" (humanloop.com) | Platform sunset September 8th, 2025; billing stopped July 30th, 2025 (migration guide) | No live product |
Three more tools changed their name or terms without a change of owner on the pages we read, and one left the category:
- Datadog now presents the product as Agent Observability, while its docs path still reads
/llm_observability. No date given. - PostHog titles its docs AI Observability, and the old
/docs/llm-analyticsURL resolved to/docs/ai-observabilitywhen we checked. No date given. - Lunary now requires payment for new cloud workspaces. Its pricing page says "This payment requirement applies only to new cloud workspaces created after this change." No date given.
- Patronus AI's homepage now describes research into tasks and environments for reinforcement learning, and presents no evaluation or observability product or pricing. No date given.
Why the owner matters more than the feature list#
Feature comparisons age in months. Ownership changes what the product is for. Read what each acquirer said it bought the tool to do:
- OpenAI said it will integrate Promptfoo's technology into OpenAI Frontier, "our platform for building and operating AI coworkers". It also said it will keep building the open-source project, and the Promptfoo README says "Promptfoo remains open source and MIT licensed."
- Cisco said Galileo would strengthen its Splunk Observability portfolio. On October 18, 2026,
galileo.ai/pricingredirected to a Splunk pricing page for Splunk Agent Observability. - ServiceNow wrote that, through "the recently completed acquisition of Traceloop", its AI Control Tower now observes AI agent behavior at runtime.
- ClickHouse: the Langfuse team's own FAQ answer to "Is Langfuse still open source?" was "Yes. No licensing changes planned." Langfuse's license file now names ClickHouse, Inc. as the copyright holder.
- Mintlify: Helicone's founders wrote that the service will "remain live for the foreseeable future in maintenance mode", with security updates, new models and bug fixes still shipping.
The pattern is the Stratechery one: tracing and evals are becoming a feature of a bigger platform rather than a standalone business. The winners are the platforms that already hold something you cannot easily move - your analytics warehouse (ClickHouse), your model provider (OpenAI), your enterprise monitoring contract (Cisco's Splunk), your IT workflow system (ServiceNow), your GPU cloud (CoreWeave). The tool you chose becomes the on-ramp to that platform. The second-order effect is that roadmaps start to follow the parent: an eval tool inside a model lab is pulled toward that lab's agent platform, and a tracer inside a monitoring suite is pulled toward that suite's pricing unit. Neither is bad on its own. Both are reasons not to let a vendor become the only place your traces and test cases live.
Humanloop is the case that makes the rest concrete. Its migration guide says "the Humanloop platform will be sunset on September 8th, 2025", and "All billing has stopped since July 30th, 2025." Teams whose prompts, datasets and eval history lived only inside it had a deadline to export them.
Portability checklist#
Portability comes down to two questions: can you point your existing instrumentation at a different backend, and could you run the tool yourself if the hosted version changed? Here is what each tool's own pages say, from the directory. "Not stated" means the pages we read did not say; it is not evidence that the feature is missing.
| Tool | OpenTelemetry / OTLP ingest | Self-host | License |
|---|---|---|---|
| Langfuse | Yes, OTLP endpoint at /api/public/otel | Core features free to self-host | MIT Expat outside ee/ |
| LangSmith | Yes, via an OTel tracer and exporter | Enterprise add-on | Not stated |
| Braintrust | Yes, OTLP traces and logs | Enterprise plan only | Not stated |
| Arize Phoenix | Built on OpenTelemetry | Yes | Elastic License 2.0 (ELv2) |
| Arize AX | Listed as "OpenTelemetry compliant" | SaaS or self-hosted | Not stated |
| W&B Weave | Yes, OTLP endpoint | Not stated | Apache 2.0 (Weave repository) |
| Helicone | Not stated | Yes | Apache 2.0 |
| Datadog Agent Observability | Yes, OTel 1.37+ GenAI conventions or OpenInference | Not stated | Not stated (commercial SaaS) |
| Honeycomb | Yes | Not stated | Not stated (commercial SaaS) |
| Pydantic Logfire | Yes, all OTel signals | Enterprise license; platform is closed source | MIT (SDK repository) |
| Opik (Comet) | Native OTel support | Yes, without a commercial license | Apache 2.0 |
| Laminar | OTLP trace ingestion | Yes | Apache 2.0 |
| Lunary | Not stated | Community Edition free; Enterprise Edition paid | Not stated |
| HoneyHive | OTLP or JSON | Enterprise plan | Not stated |
| Galileo | Not stated | Cloud and self-hosted deployments (Splunk page) | Not stated (commercial) |
| PostHog AI Observability | Not stated | Open-source deployments, no support or guarantees | MIT Expat except ee/ |
| Traceloop / OpenLLMetry | Built on OpenTelemetry | On-prem deployment | Apache 2.0 (OpenLLMetry) |
| AgentOps | Not stated | Yes | MIT |
| Confident AI / DeepEval | Not stated | Dedicated on-prem (Enterprise) | Apache 2.0 (DeepEval) |
| Promptfoo | Not stated | Run locally or self-host | MIT |
What to do with that table, in order:
- Instrument once, with OpenTelemetry. Thirteen of the twenty tools above state OTel or OTLP ingest on their own pages, including three of the acquired ones (Langfuse, Weave and Traceloop's OpenLLMetry). If your spans follow the OpenTelemetry GenAI semantic conventions, which Datadog's docs cite by version, changing backends becomes a configuration change rather than a rewrite. The standard OTel SDK variables are
OTEL_EXPORTER_OTLP_ENDPOINTandOTEL_EXPORTER_OTLP_HEADERS(OTel exporter configuration); keep the vendor URL and key there, not in code. For a worked setup, see Claude API OpenTelemetry tracing. - Keep a copy of your traces you control. Run a local or self-hosted collector in front of the vendor so you can fan out to a second destination or a file sink. Local OpenTelemetry traces as agent receipts shows the local half of that.
- Keep evals in your repository. Test cases, scorers and thresholds belong in version control next to the code they test. Promptfoo's YAML configs already work this way, which is why an ownership change there is a smaller event for users than a hosted-only eval suite disappearing. AI agent eval tools compared covers which tools run as a CI step, and why agent evals need baseline receipts covers what to store.
- Prefer a license you could fork. MIT and Apache 2.0 give you an exit if the hosted product changes. Phoenix is self-hostable under the Elastic License 2.0, a different license from MIT or Apache 2.0, so read its terms before you plan around it.
- Watch the pricing URL. Two of the changes in this post were visible first as redirects:
wandb.ai/site/pricingresolved to CoreWeave's Forge pricing page, andgalileo.ai/pricingto Splunk. A redirect is the cheapest early warning you will get.
If you are on an acquired tool today#
Langfuse. Nothing forces a move. Langfuse said Cloud keeps running with "Same product, same endpoints, same contracts", the core stays MIT, and it accepts OTLP. Keep your instrumentation vendor-neutral and the exit stays open. Our Langfuse vs Braintrust vs Helicone comparison has the current plans and self-host options.
Promptfoo. The CLI and configs are yours: MIT licensed, and run locally or self-hosted. Pin the version you depend on and watch where new features land, since OpenAI said the enterprise capabilities go into Frontier.
Galileo. Pricing is now on Splunk's page: "Starting at just $100 per month per 1.2 million spans annually, billed annually". The pages we read do not state OTLP ingest, so if you need an exit, plan it around your own instrumentation rather than an export feature.
Traceloop. OpenLLMetry is Apache 2.0 and built on OpenTelemetry, so your instrumentation is already portable. If you send spans to Traceloop's hosted platform, you can repoint the exporter without touching application code.
W&B Weave. Weave accepts OTLP and its repository is Apache 2.0. The practical change is commercial: plans are priced on CoreWeave's Forge page.
Helicone. Maintenance mode means existing integrations keep working, with security updates and new models still promised. The pages we read do not state OTLP ingest, so plan a move around your own instrumentation. If you route requests through Helicone as a proxy, the LLM gateways directory lists the routing alternatives.
Humanloop. The platform is gone. The migration guide is the record of what was offered.
Tools with no ownership change shown#
As of October 18, 2026, the pages we read showed no acquisition, shutdown or rename notice for LangSmith, Braintrust, Arize Phoenix and Arize AX, Honeycomb, Pydantic Logfire, Opik, Laminar, HoneyHive, AgentOps and Confident AI / DeepEval. That is a statement about those pages on that day, not a claim that any of them is independent or will stay that way. The portability checklist applies to them just the same.
What people are actually saying#
- Consolidation is the story, not the deal. On the Hacker News thread about ClickHouse acquiring Langfuse, one commenter framed it as part of a wider fight over which general-purpose data vendor gets to store and query all of a company's observability and business data.
- The counter-case: you may not need a specialist. In the same thread, a commenter said that Sentry, PostHog and others now do the same thing and that they had moved from Langfuse to Sentry. Another asked what a specialized LLM tracer offers over a general tracing product such as Logfire.
- Loyal users are hopeful. Others called Langfuse their favorite LLM observability tool and hoped the acquisition makes it better rather than worse. One commenter said they would move away, citing US data-access laws.
- Open-source promises are what users asked about first. On the Promptfoo thread, co-founder Michael D'Angelo answered that question before anyone asked it: the repo stays public under the same license, multiple providers stay supported, and PR reviews and releases continue. A long-time user replied hoping the project stays maintained.
- Acquisitions open gaps. About a week later, an Ask HN post used the Promptfoo deal to ask what MCP server authors use for testing in CI, arguing Promptfoo did not handle MCP transport or tool-schema validation.
FAQ#
Which LLM observability tools were acquired in 2025 and 2026?#
From the vendors' own pages: Langfuse (ClickHouse, January 2026), Promptfoo (OpenAI, announced March 9, 2026), Galileo (Cisco, completed May 22, 2026), Traceloop (ServiceNow, March 2026 announcement, completed by May 5, 2026), Helicone (Mintlify, March 3, 2026) and Weights & Biases, which makes Weave (CoreWeave, completed May 5, 2025). Humanloop's team is joining Anthropic and its platform was sunset on September 8, 2025.
Is Langfuse still open source after the ClickHouse acquisition?#
Yes. Code outside the ee/ directories is under the MIT Expat license, core features are free to self-host, and the team's announcement FAQ says "No licensing changes planned."
Is Promptfoo still open source after joining OpenAI?#
Yes. The README says "Promptfoo remains open source and MIT licensed." OpenAI said it will keep building the open-source project while integrating the technology into OpenAI Frontier.
How do I avoid lock-in with an LLM observability tool?#
Instrument with OpenTelemetry and send spans over OTLP, keep a copy of traces in a collector you control, keep eval datasets and scorers in your repository, and prefer tools with an MIT or Apache 2.0 license you could self-host. The full per-tool table is in the LLM observability directory.
Continue Reading#
- Langfuse vs Braintrust vs Helicone: Agent Tracing Tools in 2026 - plans, self-hosting and OTLP for the three most-compared tools
- AI Agent Eval Tools Compared: Braintrust vs Promptfoo vs DeepEval - the eval layer on its own, including CI runners
- Local OpenTelemetry Traces Are Agent Receipts - keep a copy of every trace before it reaches a vendor
- Claude API OpenTelemetry Tracing: How to Set It Up - vendor-neutral spans from the model provider side
- Self-Hosted vs Managed AI Gateways: A Decision Guide - the routing layer that often sits next to tracing
- LLM Observability and Evaluation Tools Compared - all 20 tools, who changed hands, every value quoted from the vendor's own page
Sources#
- ClickHouse acquires Langfuse (Langfuse announcement, Jan 16, 2026)
- Langfuse README
- OpenAI to acquire Promptfoo (March 9, 2026)
- Promptfoo is joining OpenAI (Promptfoo blog)
- Promptfoo README
- Cisco Announces Intent to Acquire Galileo (updated May 22, 2026)
- Galileo homepage
- Splunk pricing: Agent Observability
- Traceloop is joining ServiceNow (Traceloop blog, March 2026)
- ServiceNow expands AI Control Tower (May 5, 2026)
- Traceloop pricing
- CoreWeave Completes Acquisition of Weights & Biases (May 5, 2025)
- CoreWeave Forge pricing
- Helicone is joining Mintlify (Mar 3, 2026)
- Helicone pricing
- Humanloop homepage
- Migrating from Humanloop
- Patronus AI homepage
- Datadog Agent Observability docs
- PostHog AI Observability docs
- Lunary pricing
- OpenTelemetry OTLP exporter configuration
- Hacker News: ClickHouse acquires Langfuse
- Hacker News: Promptfoo Is Joining OpenAI
- Ask HN: With Promptfoo acquired by OpenAI, what are MCP devs using for testing?
Get the next deep dive like this in your inbox
One email a week on LLM Observability and the rest of the AI dev stack. Free.
Read next on AI coding tools
Langfuse vs Braintrust vs Helicone: Agent Tracing Tools in 2026
Langfuse vs Braintrust vs Helicone, plus LangSmith and Arize Phoenix: 2026 pricing tiers, self-host options, agent-trace features, and OpenTelemetry support.
14 min readAI Agent Eval Tools Compared: Braintrust vs Promptfoo vs DeepEval
AI agent eval tools compared - Braintrust, Langfuse, Promptfoo, DeepEval, Ragas, and OpenAI Evals on offline vs online evals, LLM-as-judge, CI, and datasets.
9 min readLocal OpenTelemetry Traces Are Agent Receipts
Local OpenTelemetry traces for AI agents: use OTLP and Vercel AI SDK telemetry to get receipts for model calls, tool use, latency, errors and cost.
9 min readNew here? Start with
Technical content at the intersection of AI and development. Building with AI agents, Claude Code, and modern dev tools - then showing you exactly how it works.



