A Langfuse alternative built for debugging, not just dashboards

Q: How is Langprobe different from Langfuse?

Both are open-source, self-hosted LLM observability platforms. Langfuse is a mature, MIT-licensed dashboard for tracing, evals, and prompt management. Langprobe adds a debugging layer Langfuse doesn't have: replay (edit a prompt, model, or tool config on a broken run, re-run it against a real model, and diff span by span with a determinism verdict) and a token-budgeted agent-native MCP surface so an agent can debug an agent.

Q: Is Langprobe open source like Langfuse?

Yes. Langprobe is Apache-2.0 open source and self-hosted; Langfuse is MIT open source and self-hosted. Both keep your trace data in your own infrastructure. The difference is the feature set, not the openness.

Q: Can I use Langprobe with OpenTelemetry like Langfuse?

Yes. Langprobe ingests OpenTelemetry OTLP/HTTP and OpenInference instrumentors for CrewAI, DSPy, Pydantic AI, OpenAI Agents, LlamaIndex, LangChain, and bare providers — the same open-standard instrumentation approach Langfuse supports.

Langfuse is a strong, mature open-source observability dashboard. Langprobe shares the same open, self-hosted foundation and adds a layer Langfuse doesn't have: replay and an agent-native MCP debug loop.

Last updated: July 4, 2026

TL;DR

Langfuse and Langprobe are both open-source and self-hosted — this isn't an open-vs-closed comparison.
Langfuse excels as a tracing, eval, and prompt-management dashboard with a large, mature community.
Langprobe adds replay (edit → re-run against a real model → span-level diff + determinism verdict) and an agent-native REST/MCP surface — a debug loop, not just observability.
Both ingest OpenTelemetry OTLP/HTTP; migrating traces is straightforward.

Langprobe vs Langfuse at a glance

Dimension	Langprobe	Langfuse
License	Apache-2.0 open source	MIT open source
Self-hosting	Yes — docker compose	Yes — docker compose
Tracing & spans	Yes	Yes
Evals & datasets	Yes	Yes
Prompt management	Focused on debugging	Yes — mature
Community maturity	Newer	Large, established
Replay (edit & re-run a broken run)	Yes — span diff + determinism	No
Agent-native MCP debug loop	Yes — token-budgeted find/read/replay/diff	Docs MCP + API; no replay loop
OpenTelemetry ingestion	Yes — OTLP/HTTP	Yes — OTLP/HTTP

Langfuse is a well-built product with a large community — if a mature dashboard with prompt management is your priority, it's an excellent choice. This page is about when a replay-first debugger fits better. See also LangSmith vs Langfuse.

Where Langprobe pulls ahead

Replay closes the loop

A dashboard tells you a run failed. Langprobe lets you fix it in place: edit the prompt, model, or tool config, re-run against a real model, and get a span-level diff plus a determinism verdict — so you know a fix is real, not a lucky sample. Observability ends at "what happened"; Langprobe continues to "what happens if I change this."

Agents debug agents

Langprobe's reads are token-budgeted, LLM-legible projections over REST and MCP. A 48k-token trace becomes a ~2k-token salient slice, so a coding agent can find → read → replay → diff without a human in the loop. The whole debug cycle is available as API calls.

Choose the right tool

Choose Langprobe if…

You want to edit and re-run broken runs, not just view them.
You need a determinism verdict to trust a fix.
You want agents/CI to drive debugging over MCP.
Replay and an agent-native loop matter more than prompt-management breadth.

Choose Langfuse if…

You want a mature, battle-tested observability dashboard today.
Prompt management and a large plugin/community ecosystem are priorities.
You're standardizing on a widely adopted MIT-licensed platform.
You don't need replay or an agent-native debug loop yet.

Frequently asked questions

How is Langprobe different from Langfuse?

Both are open-source, self-hosted observability platforms. Langfuse is a mature MIT-licensed dashboard for tracing, evals, and prompt management. Langprobe adds replay (edit → re-run against a real model → span diff + determinism verdict) and a token-budgeted agent-native MCP surface so an agent can debug an agent.

Is Langprobe open source like Langfuse?

Yes — Langprobe is Apache-2.0, Langfuse is MIT. Both are self-hosted and keep trace data in your infrastructure. The difference is the feature set, not the openness.

Can I use Langprobe with OpenTelemetry like Langfuse?

Yes. Langprobe ingests OTLP/HTTP and OpenInference instrumentors for CrewAI, DSPy, Pydantic AI, OpenAI Agents, LlamaIndex, LangChain, and bare providers.

Try replay on your own traces

Open source, in your VPC, Apache-2.0. Free to start.

Start self-hosting Star on GitHub