Humanloop Alternatives: Where to Move Prompts, Evals and Logs
November 17, 2026

Humanloop is gone, so every Humanloop alternative is a migration target, not a comparison. If you want an open-source prompt registry you can self-host, move to Langfuse. If product managers and domain experts edit prompts without touching code, move to PromptLayer. If you build on LangChain, use LangSmith. If evaluation was the main reason you used Humanloop, Braintrust is the closest fit. If you need a permissive licence on the whole platform, look at Opik. Agenta is the other open-source option, though it has repositioned around agents.
Here's what happened. The Humanloop homepage now says the Humanloop team is joining Anthropic and that the platform is being sunset. Its migration guide gives the dates: billing stopped on July 30, 2025, and the platform was sunset on September 8, 2025, the same day accounts and all associated data were permanently deleted. Humanloop's own guide names two places to go, Langfuse and Braintrust, and says there are now many good tools for evaluation and prompt management.
So this guide is about rebuilding. It covers the three things Humanloop held for you: prompts with their versions and deployments, evaluations, and logs.
Quick Comparison
| Tool | Prompt registry and versioning | Evals | Logs and traces | Self-host | Best for |
|---|---|---|---|---|---|
| Langfuse | Yes, with labels and caching | Yes | Yes | Yes, MIT core | Default open-source landing spot |
| PromptLayer | Yes, visual CMS, no-code editing | Yes | Yes | Not covered here | Non-engineers editing prompts |
| LangSmith | Yes, Staging and Production environments | Yes | Yes | Enterprise only | LangChain and LangGraph teams |
| Braintrust | Yes | Yes, its core strength | Yes | BYOC on Enterprise | Eval-driven teams |
| Opik | Yes, prompt library | Yes | Yes | Yes, Apache-2.0 | Permissive licence end to end |
| Agenta | Yes, prompt and deployment registry | Yes | Yes | Yes, MIT core | Open-source, agent-focused teams |
What You Need to Rebuild After Humanloop
Humanloop bundled several jobs into one product. Before you pick a replacement, list which ones you actually used:
- A prompt registry. Prompts stored outside your code, with full version history, so you could change a prompt without a redeploy.
- Deployments. Pointing an environment, like production, at a specific prompt version, and rolling back.
- An editor for non-engineers. Product and domain experts tweaking prompts in a UI.
- Evaluations. Datasets, evaluators (LLM, code or human), and runs comparing versions.
- Logs. Every production call, linked to the prompt version that produced it.
If you exported before the shutdown, you have the raw material. The migration guide points to the humanloop-export tool, a Python script that wrote Files (prompts, tools, datasets, flows, agents and evaluators), Versions, Deployments, Logs and Evaluations out as JSON and JSONL. None of the vendors below documents a one-click Humanloop importer that we could find. Plan on a short script that reads your export and calls the new tool's SDK.
If you didn't export, the data is gone. Per Humanloop's guide, it was permanently deleted on September 8, 2025. Your prompts likely still exist in your code, your git history or your application's fallbacks. Start there.
The Best Humanloop Alternatives
Langfuse: the default open-source landing spot
Best for: teams that want prompts, evals and logs in one tool they can self-host, and the option Humanloop itself named first.
Langfuse prompt management stores prompts centrally, versions them, and serves them through the SDK with server-side and client-side caching, so fetching a prompt doesn't add latency. Labels like production point at a version, which maps cleanly to Humanloop deployments. A playground lets you jump from a bad trace straight into editing the prompt behind it. Tracing, datasets, LLM-as-a-judge evals and human annotation cover the rest.
The Langfuse repo is MIT outside its ee directories. Since January 2026, Langfuse has been part of ClickHouse.
Migration fit: strong. Humanloop prompt versions become Langfuse prompt versions, deployments become labels, and datasets map to datasets. Logs are the hardest part: you'd backfill them as traces, or simply start fresh.
Watch out for: self-hosting means running Postgres, ClickHouse, Redis and blob storage. Our Langfuse alternatives guide covers the trade-offs.
PromptLayer: prompt management for mixed teams
Best for: teams where product, marketing or content people did most of the prompt editing in Humanloop.
PromptLayer calls itself "the collaboration layer for AI engineering teams." It has a visual prompt CMS and registry, diffs, comments and rollback, release to dev or prod, and gradual A/B tests between prompt versions. Domain experts can edit and test prompts from the dashboard. Evals cover dataset-backed tests, human and AI graders, regression tests and backtests against historical logs, and production logs link back to prompt versions.
Migration fit: the closest match to how non-engineers used Humanloop's editor.
Watch out for: its docs show datasets and evaluations moving into a newer Tables feature, so build new evals there rather than in the legacy views.
LangSmith: prompts with environments, for LangChain teams
Best for: teams whose apps already run on LangChain or LangGraph.
LangSmith's prompt management uses commits with diffs, commit tags, and reserved Staging and Production environments you promote commits between. Prompt owners control who can promote or delete, and a webhook can fire on every prompt commit, which is handy for syncing prompts into your repo. Datasets, evaluators and tracing sit alongside.
Migration fit: Humanloop deployments map to LangSmith environments almost one to one.
Watch out for: LangChain's pricing charges per seat plus traces, and self-hosting is an Enterprise option. Our LangSmith alternatives guide covers what to do if seats are the problem.
Braintrust: if evals were the point
Best for: teams that used Humanloop mainly to run evaluations before shipping prompt changes.
Braintrust is the other tool Humanloop's guide recommends. It's built around evals: datasets, scorers, experiments that compare prompts and models side by side, and Loop, an AI agent that helps write prompts, scorers and datasets. Prompts live in the platform too, and production traces can become eval datasets. Its pricing page lists unlimited users on every plan.
Migration fit: strongest for your evaluation history and datasets.
Watch out for: new organizations can't self-host; Braintrust offers bring-your-own-cloud on its Enterprise plan instead. For a full breakdown of Braintrust's own competitors, see our Braintrust alternatives guide.
Opik: open source down to the backend
Best for: teams that want a self-hosted replacement with no licence questions.
Opik, built by Comet, is Apache-2.0 across the backend, web app, tracing, datasets, experiments, prompt management and online evaluation. Its prompt library versions prompts and parameters in a playground and lets you deploy the versions that work. It also has a prompt optimizer and a PyTest integration for running evals in CI.
Migration fit: good for teams that want Humanloop's prompts-plus-evals shape without a vendor contract.
Watch out for: a smaller integration ecosystem than Langfuse or LangSmith.
Agenta: open source, now agent-first
Best for: teams that want an open-source registry and expect to build agents next.
Agenta is MIT outside its ee directories and self-hostable. Its docs cover a prompt and deployment registry, prompt versioning, folders, evaluations from the UI or SDK, and webhooks or GitHub Actions that fire when you deploy a prompt. The project has repositioned, though: its README now leads with "the open-source workspace for your agents and your team," building AI coworkers you talk to in Slack, WhatsApp or Telegram.
Migration fit: the prompt registry and evals still map to Humanloop's model.
Watch out for: product focus has shifted to agents, so check that the prompt-management features you need are still on the roadmap.
One Former Option to Handle With Care
Vellum has changed direction. It used to come up alongside Humanloop as a prompt and workflow platform. The Vellum homepage now sells "a personal AI assistant for your busywork," with an open-source assistant you can run on your own hardware. Its developer documentation still describes Prompts, Workflows, Deployments and Evaluation, but the platform changelog's latest monthly entry is February 2026. Confirm with Vellum directly before you move a production prompt registry there.
How to Choose a Humanloop Alternative
| What you used Humanloop for | Move to | Why |
|---|---|---|
| Prompt registry plus self-hosting | Langfuse, Opik, Agenta | Open-source registries you can run yourself |
| Non-engineers editing prompts | PromptLayer | Visual CMS and no-code editing |
| Environments and promotion | LangSmith, Langfuse | Staging/Production environments or labels |
| Evals before every release | Braintrust, Opik | Eval-first workflows and CI hooks |
| You build on LangChain | LangSmith | Native tracing and prompt hub |
A practical migration order:
- Prompts first. Recreate your production prompts and their current deployed versions. Older versions are nice to have, not essential.
- Point production at the new registry. Swap the Humanloop SDK call for the new tool's fetch-prompt call, and keep a hard-coded fallback in your code. Watch template syntax: if your prompts used double-brace variables like
{{name}}, check how the new tool expects them. - Then datasets and evaluators. Rebuild your top two or three evaluators and rerun them on the migrated dataset to confirm scores look sensible.
- Logs last, or never. Old logs matter for audits and backtests. For most teams, starting fresh traces on the new tool is fine.
Where Taku Fits
All of these tools help teams manage the prompts behind AI apps they build. Plenty of people never build one; they want to use AI apps someone else has already built and tuned. Taku isn't a prompt-management platform, and it doesn't replace Langfuse, PromptLayer or any tool here. That's where Taku comes in.
Taku Apps and Sites, built by the best AI minds. Find one, run it, and keep it in your Mac Dock. Taku is in Beta, and the Mac app is available now.
Key Points
- Humanloop's team is joining Anthropic. The platform was sunset on September 8, 2025, and Humanloop says account data was permanently deleted the same day.
- Humanloop's own migration guide recommends Langfuse and Braintrust.
- Langfuse, Opik and Agenta are self-hostable; LangSmith is self-hosted only on Enterprise, and Braintrust offers BYOC instead for new organizations.
- PromptLayer is the closest match for non-engineers who edited prompts in Humanloop's UI.
- Vellum's homepage now sells a personal AI assistant, so confirm its platform status before migrating there.
- Migrate prompts first, then evals, and treat old logs as optional.
FAQ
Is Humanloop shut down?
Yes. Humanloop's migration guide says the platform was sunset on September 8, 2025, and its homepage says the team is joining Anthropic.
Can I still export my data from Humanloop?
No. The guide says accounts and all associated data were permanently deleted on September 8, 2025. If you ran the humanloop-export tool before then, you have JSON and JSONL files you can script into a new tool.
What did Humanloop recommend migrating to?
Its migration guide names Langfuse and Braintrust, and notes there are now many good tools for AI evaluation and prompt management.
What's the best open-source Humanloop alternative?
Langfuse is the most common choice because it covers prompts, evals and traces and self-hosts without a contract. Opik is the pick if you want Apache-2.0 across the whole platform.
Did Anthropic launch a replacement for Humanloop?
Not that Humanloop's announcement says. It describes the team joining Anthropic and the platform sunsetting, not a successor product.