Xander Song

OSS @ Arize AI
Building @ArizePhoenix

[Mastra @mastra]

it takes one to know one

tsai builders, that is

@seldo from @arizeai and @denise_teng25 from @GradientVC are emceeing tomorrow's TSAI conf

livestream starts 9:35am PT

https://t.co/zhd3xfAJlX

[Arize AI @arizeai]

Solid WORKSHOP happening tomorrow with @schavalii covering when to use reasoning, CoT, and explanations for LLM-as-a-Judge. RSVP: https://t.co/kErnTwcl26

[Xander Song @axiomofjoy]

Super excited for the work that Yasmine has been doing here! Check out the much improved experiment views 🧪🤓

[arize-phoenix @ArizePhoenix]

Comparing experiments in Phoenix just got a lot easier.

The new List View helps you quickly scan results with per-example metrics, while the Metrics View gives you a high-level look at how changes impact cost, latency, tokens, and more.

Upgrade to v11.32.1 to start exploring.

[John Gilhuly @JohnGilhuly]

Connect OTEL tracing between MCP clients and servers 🤝

Huge shoutout to @adrianfcole and Anuraag Agrawal for contributing this feature.

We've just released a new instrumentor in @ArizePhoenix's OpenInference OSS library. This instrumentor allows you to propagate @opentelemetry context between MCP clients and servers to create a single unified trace of the full run of your application.

To use it, instrument your MCP client and server using OpenTelemetry, and call the new openinference-instrumentation-mcp library. Full example in the comments below.

Hope you enjoy this new way to get more visibility into your MCP applications!

@mikeldking @axiomofjoy @arizeai

[arize-phoenix @ArizePhoenix]

🎉 5000 Stars and Counting... 🎉

We're celebrating Phoenix reaching 5000 stars on GitHub!

This milestone underscores the growing demand for robust, open-source tools that tackle the complexities of AI and LLM development

If you're new here, Phoenix allows AI engineers and data scientists to quickly visualize their data, evaluate performance, track down issues, and export data to improve

We've just released a video celebrating this milestone and highlighting the impact of Phoenix

Check it out and discover how Phoenix can revolutionize your AI workflows https://t.co/rsKeegSVwl

[Xander Song @axiomofjoy]

@samuelcolvin @pydantic Glad to see you and your team in there! OTel events vs. attributes is a must-watch 🌶️

[Xander Song @axiomofjoy]

@matthausk @dltHub @continuedev Interesting, are you mainly using MCP features like tools, or also features like prompts and resources?

[Xander Song @axiomofjoy]

Who's building with model context protocol? Curious if we should build instrumentation for it.

[Xander Song @axiomofjoy]

@mikeldking That's so cool

[John Gilhuly @JohnGilhuly]

We've added DAY 0 support of @GoogleAI 's Gemini 2.0 🤖 🚀 to @ArizePhoenix . Totally open-source, totally free

@Google @OfficialLoganK

[Xander Song @axiomofjoy]

As always, Phoenix is open-source and built with ❤️

If you found this useful, leave us a ⭐ on GitHub

Links in the bio of @ArizePhoenix

🧵 6/6

[Xander Song @axiomofjoy]

👥💬 Sessions are out in @ArizePhoenix 7.0

View the top-level messages of a conversation before diving into details

Demo w/ @llama_index 🦙

@jerryjliu0

🧵 1/6

[Xander Song @axiomofjoy]

🛠️ In your application code, group traces into sessions using our context managers

🔑 You have control of the session ID, which can be an arbitrary string

👥 You can add user IDs for the participants in the conversation, which will show up on each trace in the UI

🧵 5/6

[Hamel Husain @HamelHusain]

I always get asked what my favorite **Open Source** eval tool is. The most impressive one thus far to me is Phoenix

Very permissive license as well.

https://t.co/s5FAUkVzn3

[Xander Song @axiomofjoy]

As always, Phoenix is fully open-source. Links in the bio of @ArizePhoenix. If you like what you see, please leave us a⭐️on GitHub🙏

🧵7/7

[Xander Song @axiomofjoy]

🛝Prompt Playground is out in Phoenix 6.0

Makes it easy to iterate on prompts, replay spans, and run experiments

🧵1/7

[Xander Song @axiomofjoy]

🧪Datasets and Experiments

- 📊Run up to four configurations over an entire dataset at once
- 🤖Automatically record experiments, then evaluate using LLM-as-a-judge and code evaluators

🧵6/7