Marc Klingen
@MarcKlingen
Followers
1,116
Following
915
Media
107
Statuses
683
building @langfuse (YC W23) – open source llm engineering platform
Berlin
Joined July 2015
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
#BAMESISSHOWCASEINCALGARY
• 197805 Tweets
HAPPY MOMENTS WITH BAMBAM
• 134774 Tweets
Nebraska
• 132882 Tweets
LINGORM LAYS STAX THAILAND
• 85810 Tweets
#หนีงานที่รักพักกินLaysStax
• 79594 Tweets
Roman
• 63325 Tweets
Noroña
• 43792 Tweets
Vaticano
• 32441 Tweets
Axel
• 23074 Tweets
Saint Dr MSG Insan
• 21296 Tweets
#SaveLittleCreatures
• 19314 Tweets
大雨特別警報
• 18161 Tweets
ロッキン
• 17806 Tweets
Stanford
• 17743 Tweets
Sanatan Sanskriti
• 17458 Tweets
Andrade
• 16751 Tweets
カミガタ
• 14708 Tweets
Birds Nurturing
• 14370 Tweets
能登半島
• 14239 Tweets
From Russia with Lev
• 13402 Tweets
きらら展
• 12174 Tweets
三連休初日
• 11846 Tweets
線状降水帯
• 10081 Tweets
Pinned Tweet
We are back on
@ProductHunt
with
@langfuse
2.0
This launch week added tons of new features for prompt management, testing on datasets, an
@posthog
integration, and online evaluations.
Greatly appreciate your upvote and launch feedback on PH!
😻 We're back on
@ProductHunt
- where we originally launched
@langfuse
last year.
🆙 Check out our launch and lend us your support:
👇 Marc is introducing Langfuse 2.0 and running through all new features in the video below
6
5
58
2
0
23
I'm super excited to share this news 🎉
Humbled to work with such an amazing community and set of investors to build the leading open source LLM analytics platform on top of our existing observability stack.
We’ve raised a $4M seed round led by
@lightspeedvp
and
@lafamigliaVC
with participation from
@ycombinator
.
11
17
123
25
9
173
Today's
@ycombinator
demo day 🧡
It's been a journey that changed the trajectory of the team, still can't believe that
@gustaf
selected us 🙏
Great job everyone presenting 👏 and thanks to
@rawert
who captured the moment
4
1
87
Proud to launch 🪢
@langfuse
today – open-source product analytics for LLM apps!
We help teams building with LLMs turn production data into insight. Quick overview: 1/x
Congrats on the launch,
@rawert
,
@marcklingen
&
@maxdeichmann
!
@Langfuse
(YC W23) is building open-source product analytics for LLM apps, helping companies to track and analyze the quality, cost, and latency of LLMs, chains & agents in granular detail.
2
2
35
9
10
84
Self-hosting Langfuse just got easier
Just pull the latest Docker image and all pending migrations are applied automatically ✨
See changelog for full details:
0
2
10
1
5
29
Launch early, launch often – thx for 🏆
#1
, users & PH community 🙏
Great to see many new people reach out and discover the project!
Repo:
Launch:
Discord:
4
1
25
It is launch day again, excited to share Langfuse with more users. Help us get to
#1
🚀
cc
@maxdeichmann
@rawert
3
3
24
✨ Friday night project: new changelog for
@langfuse
Building it and seeing what we've shipped each week feels great!
7
3
24
3 weeks in SF: YC reunion first weekend, quality time with users, Fleet Week on last day. Life's good
Until next time 👋
0
0
23
Just added
@langfuse
to a project in 19 seconds and you can too 🚀
Launching: Drop-in replacement of OpenAI SDK to get started with Langfuse super-⚡️
1
2
15
2
1
25
Lots of cool engineering problems to solve, come build
@langfuse
with us 🤗
We're hiring a founding engineer to join us in building out OSS LLM Observability
1
6
23
0
3
19
Currently working on a lot of integrations to make it easier to get started with
@langfuse
, excited about the one with Langflow (
@logspace_ai
) ✨
We are very excited to launch our integration with langflow by
@logspace_ai
today. More below in the thread:
1
3
11
0
2
17
Our super secret (open-source) prompt leaked 🫣🚨
Thanks for the interesting example
@charles_irl
!
Checkout the public trace in
@langfuse
to learn more about the arcane tactics applied to our docs q&a chatbot:
Having worked in ML prior to the rise of LLMs, I think I understand the cracking culture of the 70s and 80s a lot better.
It's just _fun_ to use arcane technical knowledge and wits to get a complex system to misbehave. And, for now, there are no real consequences.
0
0
13
2
2
15
🔄 3 new integrations
Integrating with
@langfuse
in a minute if you use:
◆ Langflow (
@logspace_ai
)
◆ Flowise (
@FlowiseAI
)
◆ LiteLLM (
@LiteLLM
)
Thanks to the founders for making this happen:
@rodrigonader_
,
@henryhengzj
,
@krrish_dh
,
@ishaan_jaff
1
2
14
@langfuse
Update September
◆ Datasets (beta) to run structured experiments
◆ Model-based evaluations on production data
◆ New query engine for in-app dashboards
◆ 3 new integrations
◆ Many UI improvements
Read on 👇
1
3
12
Getting started with
@langfuse
just got easier with the new 🦜🔗
@LangChainAI
integration.
With Langfuse, users of Langchain can get nested production traces and analytics for their LLM application.
🪢 + 🦜🔗 -> 🤌
1
2
12
Team's having fun preparing some visuals for PH launch and I can't get over that the Google Glass aesthetic is probably now retro enough to catch on in Berlin
1
1
11
Super basic feature, but requested by many >> Using (filtered) production examples for fine tuning and testing just got easier
*alternatively you can also use the Python SDK or API to get data from
@langfuse
0
3
11
Excited to share what we’ve learned while building
@langfuse
Finally, I am super excited to announce the three speakers and their topics for the next LLM Meetup in Berlin 🔥
@MarcKlingen
| Which metrics to track for optimizing LLM applications
@OttoZastrow
| Beyond langchain - LLM agents for legal documents
@taivopungas
| Speeding up LLM
2
3
19
0
0
11
“Context is the product”
OpenAI's dev day updates instantly made Tab faster, cheaper, and significantly more powerful.
AI startups should be focused on novel data ingestion products, rather than simply UI improvements. Context is the product.
30
74
1K
2
2
11
Product I’m in love with that’s not LLM related: BigQuery data connector for Google Sheets
It feels like a superpower to have live input sheets based on actual product/business/CRM data
2
0
10
It’s wild how frictionless it is to experiment with different models when using
@LiteLLM
, love your work
@krrish_dh
@ishaan_jaff
👉 🚄🪢🚄🪢
1
0
9
This will be interesting, thanks for organizing
@alex_schmitti
If you’re in Berlin on Nov 7, come join us for an evening on how the best teams approach building their LLM based features and apps
It was about time ⏰!
Excited to bring our very own Lightspeed Generative AI Series to Berlin - finally!
On Tuesday, 7th November we will be discussing what it takes to embed LLMs effectively into SaaS applications. Seats are limited, apply now.
1
1
14
1
2
9
.
@phindsearch
is amazing
Basically a custom howto for any engineering problem as long as you put a little bit of effort into describing the problem well enough
Phind can now beat GPT-4 at programming, and does it 5x faster.
127
412
3K
1
0
8
Finally getting to work through lots of UI improvements ✨
✨ Pretty-mode for common LLM I/O
Starting with OpenAI Chat Messages
1
2
8
0
1
9
Langfuse on 🔥 this week. Great conversations with lots of new teams discovering the project via HN,
@tldrnewsletter
and now GitHub
1
0
8
Who’s built LLM-agent for CSS/Tailwind cleanup?
I feel like 20% of styles/classes could easily be deleted without any effect as it was just me trying to get an overflow right.
0
0
8
What's the best part about
@ycombinator
? Building alongside other ambitious founders in LLM/AI Dev tooling 🙏 What's even better? Many of them build open-source ❤️
Thanks
@md_rumpf
for your efforts on this!
🚀 These YC companies make building with AI 10x easier!
🧑💻 An underrated aspect of
@ycombinator
is the wealth of resources the community has created for building with LLMs & AI.
🔖 Here's a list of tools you can start using today. You don’t want to miss bookmarking this!
23
67
286
1
0
8
@ycombinator
@gustaf
@rawert
Cringing a bit at the photo, but it's a moment I'll cherish for a long time
0
0
7
👋 Hi SF people, I’ll be back for two weeks starting Thursday. Dm if you‘d like to cowork or grab coffee/dinner
1
0
7
Update to OpenAI integration in
@langfuse
✔️ Group multiple llm calls into a single trace when they belong to the same invocation
✔️ Add scores to evaluate outputs
See docs for examples
0
1
6
🚢 Team is on a roll, excited to share what we've shipped to
@langfuse
(open-source) in July:
◆ New Trace UI
◆ Human-in-the-lop eval
◆ User-level reports
◆ Langchain integration
◆ Improved SDKs
◆ Automated token counts
Details/screenshots 🧵
2
2
7
Evaluate your production RAG traces in
@langfuse
with RAGAS; checkout the new cookbook
Great collaboration 🙌
@Shahules786
& Jithin
0
1
7
Correction,
@raycast
is still the best 🙏
Wondering why Apple has not copied the searchable clipboard history yet, it's a game changer
Product I’m in love with that’s not LLM related: BigQuery data connector for Google Sheets
It feels like a superpower to have live input sheets based on actual product/business/CRM data
2
0
10
1
0
7
Thanks for the feature
@tereza_tizkova
Appreciate what you,
@mlejva
and the team have built with the agent protocol at
@e2b_dev
👀 into how to easily instrument & analyze agents using the protocol, an integration with
@langfuse
would be 🔥 as some users analyze complex agents
You've heard about our AI agents repo. Now get ready for the list of agents SDKs. ⭐️
We welcome your input. Please share your suggestions for any modifications via PR or reply. 🙏
11
26
116
1
1
6
Was able to build this super fast thanks to:
‣ Superb
@vercel
AI SDK with convenient streaming callbacks (
@jaredpalmer
)
‣ Supabase GitHub Action to embed .mdx docs (
@kiwicopple
)
‣ Chatbot UI components supporting markdown and streaming (
@shuding_
and
@shadcn
)
1
0
6
The irony of auth errors when signing up for Azure to get OAuth credentials
1
0
5
All open source (in mdx, using Next.js & )
1
1
6
@yi_ding
100% this. We work with many teams running LLMs and Retrieval Augmented Generation apps in production. Most use JS/TS and a large share of those who use Python for technical exploration shift to JS/TS when going to production
2
1
5
And thanks again to
@gustaf
and
@dflieb
and our friends at
@ycombinator
for all of their support! 🙏
We're excited to work with many YC founders building production-grade LLM apps!
1
0
5
@alex_schmitti
@langfuse
@maxdeichmann
@rawert
@Lightspeed
Thanks for your excitement, vision and support. I am super pumped to work with you and team Lightspeed 🚀
0
0
5
Excited to see what our users will build with GPT-4 Turbo 128k and GPT-4 fine tuning 🚀
Until now many worked hard to reduce costs and dynamically used smaller models for specialized use cases.
1
0
4
For more details check out:
Overview:
Repo:
Docs:
1
0
4
.
@ShortbreadAI
is impressive and a lot of fun – even though I’m not into comics. It’s so effortless.
Great job
@PFengjiao
1
0
5
Love to have blog/docs/landingpage all in one project, makes it super simple to e.g. reuse changelog on landingpage
0
0
4
More relevant than ever:
“The data moat is how you populate the context window. You’d better be really goddamn good at fetching the right data to stuff into that 100k-char window. Because that’s the only way to affect the quality of the LLM’s output”
0
0
4
We're here to help in case you have questions while exploring Langfuse, feel free to dm.
Excited to share further updates over the coming weeks, we're just getting started.
@rawert
,
@maxdeichmann
1
0
4
Blog post with live demo, implementation details and link to repo
1
0
3
Excited for this. One of the most requested features of teams running on Langchain. We now also support Langchain on JS/TS 🚀
🚢
⛓️🦜
Langfuse x LangChain JS/TS integration is hot off the press ☕️
1
1
8
0
0
4
Getting started with
@langfuse
just got easier with the new 🦜🔗
@LangChainAI
integration.
With Langfuse, users of Langchain can get nested production traces and analytics for their LLM application.
🪢 + 🦜🔗 -> 🤌
1
2
12
0
0
3
@gilbert
Just started listening to it, love it already! Deeply respect how consistently you are churning out super interesting episodes
0
0
2
The integration is still a bit limited in scope but we’ll expand it to supporting all features of Langfuse soon
1
0
3
2
0
3
@yi_ding
I'd attribute it to advantages of full-stack JS/TS when building for the web and superior DX of some JS/TS packages
0
0
3
@alexdanilo99
@Teddarific
Congrats
@alexdanilo99
&
@Teddarific
on the launch! AAA team & execution. Feels like ✨
0
0
2
0
0
3
Everyone can stream LLM responses from server/edge. Just talked to an early-stage team streaming responses via the real-time capabilities of
@supabase
. Might not scale but works, no state mgmt problems, apparently fast enough
0
0
3
@vietdle
@maxdeichmann
@rawert
@alex_schmitti
@lightspeedvp
Thanks for joining us on the ride since day 1 🙏 I appreciate your support and advice a lot.
If you lost track of what we shipped, check out the changelog:
0
0
3
0
0
3
Excited for what's coming in August and the overall roadmap! Talking to many users rn to go deep on some topics. Dm if you want to share your thoughts on what you really need us to build!
1
0
2
0
0
3
@TheRobHayes
@langfuse
@VoiceflowHQ
Looker is the real mvp 🙏 I wish there was a good OSS solution for embedded analytics.
We created the dashboards with our biggest users as a prototype but then realized that giving everyone access just makes us move faster.
👀
1
0
2
🌟 Human-in-the-loop Evaluation
Ditch the sheets and other processes to manually evaluate production outputs. Centralize it with your team in Langfuse and easily combine it with other ingested scores. E.g. eng can have a closer look to all negatively reviewed traces.
1
0
1
We realized that most teams building production-grade applications with LLMs suffer from a lack of tailored product analytics and monitoring.
1
0
2
Recap: we're building the open-source analytics suite for LLM applications. Helping teams make decision quickly with the right metrics and debugging tools.
Proud to launch 🪢
@langfuse
today – open-source product analytics for LLM apps!
We help teams building with LLMs turn production data into insight. Quick overview: 1/x
9
10
84
1
0
2
Try your own tactics to trick our bot while watching the traces of your experiments here:
0
0
2
@AtaOnat94
Thank you Ata, super happy to hear from you. Feel like last time we chatted you got 💍. All the best!
0
0
1
3. Collect user feedback and get to root causes quickly (initial gif)
1
0
2
@plbecker
+ just use emoji as logo, super fast “design process” and you can logo-react in every app
0
0
2
@joshm
@arcinternet
@browsercompany
There aren’t many changes to the tools you use very day that make you do things in a new way. Arc link previews is one of them. The UX is magic ✨ thank you
0
0
1
How to get started?
Docs:
Discord:
Repo:
0
0
2
@mkovarski
@langfuse
Langfuse focuses on the server-side latency of LLM apps excluding the network to the client.
We help track single llm calls (see image) and chains/agents.
Soon:
- Comparison across prompts, versions and users
- Breakdown into individual steps of a more complex LLM app
2
0
2
Why?
A user was surprised when I (a human) answered his/her question. A bot was expected lol. Shipped it the next day.
As we added a lot to the
@langfuse
docs over the last few days, building a retrieval-based chatbot finally made sense to help users explore the project.
1
0
2
✅ SSO enforcement
Chipping away at enterprise features is fun
1
0
2
Proud to launch 🪢
@langfuse
today – open-source product analytics for LLM apps!
We help teams building with LLMs turn production data into insight. Quick overview: 1/x
9
10
84
0
0
2
🧮 Automated Token Counts
Usage/cost is key to Langfuse reports as we help users monitor and optimize costs. Langfuse now automatically calculates token counts for OpenAI and Anthropic models to make reports accurate without added effort.
1
0
2
@josephdhlee
@synclyhq
@ycombinator
@500GlobalVC
@doshkim
@gustaf
@dflieb
@tchae
@jaredheyman
Congrats
@josephdhlee
and team!
1
0
2