Subscribe to Langfuse Events via Webhooks

[marcklingen]

Langfuse produces many different events. For various reasons it can be useful to subscribe to them via a webhook.

Potential events

• score:create (user feedback, eval, ...)
• trace:create
• prompt:create

Very early idea, raised by a couple of teams using Langfuse. Feel free to add your thoughts or what you'd like to see/contribute.

[khareyash05]

Hey This feature will be very useful! Will this be a Slack Webhook Url or a Discord Webhook Url?

[marcklingen]

Currently thinking about generic webhook. It is easy to then setup a Slack workflow to consume this information and print to Slack or use Zapier/Make/n8n to build a quick custom consumer in order to pipe the events to other channels.

[khareyash05]

Hey another event I can think of is whenever a new member is added/removed to/from the project
Also with webhook url we can also send an email that this person was added to the workspace by this person.

[clemra]

• 1 in [Alerts] Connect to Slack #1512

[hypernovas]

Thanks! Curious what is the status for the webhook capability? Something we were looking at is based on the cost burst, LLM API outage or latency spikes, we can publish events. This can be very useful for LLM app ops for many orgs.

[marcklingen]

We are waiting for v3 changes before adding webhook capabilities. Thereby these can happen behind async queues instead of having to be run within sync apis. More context on this here: https://github.com/orgs/langfuse/discussions/1902

[chrisshaw]

Sorry to bother twice, walking over from #1512 . Wanted to extend the very good use cases we also have that @hypernovas brought up with what I shared in my reply in the prior discussion: alerts based on sliding window aggregated scores. Responding to events from individual scores, traces, etc seems fun but not urgent. In our case, we want to know that the moderation LLM app is still moderating correctly and want to be alerted based on an aggregation of scores crossing a threshold.

[marcklingen]

thanks for the feedback and sorry for the slow response. I fully agree that alerts are very distinct from rule-based triggers on every newly created object in Langfuse. I create a new thread to track this independently: https://github.com/orgs/langfuse/discussions/3997

[ajram23]

+1 please.

[arackley-tribal]

+1 Would love to be able to use this to be notified when Langfuse is noticing errors in traces etc.

[barseghyanartur]

Copied from https://github.com/orgs/langfuse/discussions/4342

---

Feature request: Lifecycle hooks in LangFuse

Introduce lifecycle hooks in LangFuse to trigger custom actions during key events (such as prompt promotion to production).

These hooks should provide flexibility, allowing integration with external systems via webhooks.

Initial implementation could be as simple as managing that through the UI:

• Event to hook into.
• Webhook URL.
• Configuration (security tokens, retries, timeouts).
• Additional payload parameters.

On a long term, support for a reproducible configuration format for defining lifecycle hooks would be useful. Think of version-controlled deployments or "manual" environment duplications.

---

[chai3]

datasets:create
I want to run agent experiments(not prompt experiments)