@arizeai/phoenix-otel 1.0 is out. One import to trace TypeScript agents. Wrap a function, get a span — inputs, outputs, errors, span kind, all recorded automatically.

Some quality of life improvements Arize Phoenix now supports python 3.14 across all SDKs and has new REST endpoints including API / Secret rotation APIs. arize.com/docs/phoeni...

Guess how many 🌟 Phoenix has on GitHub? Yes, this meme has been all over our internal slack today. github.com/Arize-ai/ph...

We just added streamdown rendering for our LLM outputs now look and perform so well. Particularly loving the bash and code rendering 🤟

Post 1 PSA: LiteLLM versions 1.82.7 and 1.82.8 on PyPI were compromised with a credential-stealing payload. Are you using Phoenix's optional LiteLLM extra for phoenix-evals directly or through DSPy, Smolagents, or CrewAI? Did you install or upgrade to litellm 1.82.7 or 1.82.8 from PyPI? See: 🧵

Track exactly what changed in your prompts with diffs!

Arize Phoenix 13.11.0 — Adds @perplexity_ai and @togethercompute as built-in providers for benchmarking and evaluation.

Arize AI Phoenix v13.10 now supports Cerebras, Fireworks AI, Groq, and Moonshot (Kimi), as well as OpenAI's GPT 5.4 models, allowing you to compare hundreds of more models side by side for benchmarking, task evaluation, or LLM judge building. #AI #LLM #OpenSource #Observability #Evals

Anthropic's Claude Agent SDK lets you build AI agents that autonomously read files, run commands, search the web, and edit code — the same tools and agent loop that power Claude Code, now programmable in Python and TypeScript. Docs: arize.com/docs/phoeni...

This is the chart that should make every AI engineer pause. In the demo agent we evaluated: ⚪Tool selection: 100% ⚪Matches expected tool calls: 36%

Sessions are how your users actually experience your AI app — not as isolated calls, but as multi-turn conversations. Now available in the Phoenix SDKs. arize.com/docs/phoeni...

New Phoenix release: Sessions API & CLI 🔭 LLM apps are multi-turn — your observability should be too. Sessions group related traces into a conversation timeline so you can see why a chat went wrong, not just what happened on one step. arize.com/docs/phoeni...

We have a new integration: The Claude Agent SDK for typescript! www.npmjs.com/package/@ar...

Agents are writing more code than ever. But without observability, they're flying blind. Two shifts happening at once: the new authors of code and a new kind of software being written. @mikeldking believes telemetry gets us one step closer to true agent autonomy this post. arize.com/blog/closin...

Phoenix 13.0 Phoenix 13 is a major release centered around Dataset Evaluators, a new system that turns your datasets into reusable evaluation suites. This release also introduces custom model providers, OpenAI Responses API support, and dozens of Playground and experiment UX improvements.

We just published a new Get Started with Phoenix Quickstart in TypeScript 🚀 Using a Mastra multi-agent system, it walks through the workflow we recommend for improving agent apps in a way you can measure + validate:

RSVP just opened up for our workshop with @mikeldking + Hamel: maven.com/p/2c8410/au... Learn how to: ⚡ Connect Claude Code to Phoenix observability data ⚡Use CLI commands to fetch traces and debug agents ⚡Prompt AI to analyze system behavior in real-time

Go from traced runs to measurable quality metrics with @ArizePhoenix TS evals 🚀 Check it out! arize.com/docs/phoeni... arize.com/docs/phoeni... arize.com/docs/phoeni...

🔺 Use real trace data from your app runs as the basis for evaluating LLM output 🔺 Build evaluators (built-in or custom) that score outputs on correctness, relevance, and other quality criteria 🔺 Run those evaluators with Phoenix’s TypeScript eval tooling to produce structured quality metrics

Just Added: TypeScript tutorial for structured eval workflows in Phoenix 📊 Follow an end-to-end TS tutorial featuring a @langchain.bsky.social application that shows how to:

Just Added: TypeScript tutorial for structured eval workflows in Phoenix 📊 Follow an end-to-end TS tutorial featuring a @langchain.bsky.social application that shows how to:

Just Added: Python narrative tutorials for Eval workflows 📊 We’ve published python guides that walk through an Agno application with a full evals workflow: showing you how to measure & iterate on AI quality using real traces and code-driven evaluation.

New in Phoenix: End-to-End TypeScript tutorial for Datasets & Experiments 🔍📊 Datasets and experiments are essential for structured iteration—helping you move beyond ad-hoc testing and systematically validate improvements to your AI application.

New in Arize AX: Evaluator Hub 🚀 Evaluator Hub introduces reusable, versioned evaluators that can be shared across evaluation tasks, instead of being redefined per task. This makes evaluation behavior easier to reason about as tasks, datasets, and models evolve.

🚀 New OpenInference integration with @pipecat.bsky.social We released native OpenInference instrumentation for Pipecat using OTEL; enabling end-to-end observability across realtime agent pipelines. Get rich semantic traces with zero manual instrumentation in Arize Phoenix. Demo Below 🎬

We distilled best practices for using #OpenTelemetry for AI observability into an agent skill. phoenix-tracing teaches your AI coding assistant (Claude Code etc.) how to properly instrument LLM applications—setup, custom spans, production deployment. skills.sh/arize-ai/ph...

We just shipped our first agent skill: npx add-skill Arize-ai/phoenix --skill phoenix-cli arize.com/docs/phoeni... Our rationale:

01.22.2026 CLI Prompt Commands: Pipe Prompts to AI Assistants

In AI application development, rapid iteration can easily mask regressions or introduce blind spots in quality. Prompt tweaks, model swaps, or architectural changes may appear better in isolation, but without systematic evaluation, progress becomes guesswork.

Create Datasets from Traces with Span Associations Available in arize-phoenix-client 1.28.0+ (Python) and @arizeai/phoenix-client 2.0.0+ (TypeScript)

The Phoenix CLI now supports datasets, experiments, and annotations. Pull evaluation data, export experiment results, and access human feedback directly from the terminal. Datasets from the command line.

📣 the Phoenix CLI: Terminal Access for AI Coding Assistants such as @claude_code and @geminicli @arizeai/phoenix-cli is a command-line interface for retrieving trace data. It provides the same observability data available in the Phoenix UI through shell commands and file exports.

Span comes from "span of time" - That's why during that span you can have events to keep a simple log of things that happen. Phoenix now supports displaying span event attributes in the UI. span.add_event( name="model.config", attributes={} )

We released a new Python Tracing Quickstart for Phoenix! Agents don’t fail like traditional software. They can misclassify, retrieve the wrong docs, or forget context. Learn the workflows that turn agents into systems you can understand and assess: arize.com/docs/phoeni...

The agent is built in Python with @crewai.bsky.social, uses Serper for real-time search, and follows the full trace → eval → iterate loop end to end with Phoenix. Check it out: arize.com/docs/phoenix... TS version coming soon!

🔹 Trace a @crewai.bsky.social agent to capture execution flow, tool calls, and LLM calls 🔹 Define evals to score outputs and label failures 🔹 Build a dataset of failure cases as concrete data to test iterations 🔹 Run experiments to compare agent versions on the same inputs and verify improvements

We just published a new Get Started with Phoenix Quickstart It walks through the workflow we recommend for improving agent apps in a way you can measure and validate:

We just published a new Get Started with Phoenix Quickstart It walks through the workflow we recommend for improving agent apps in a way you can measure and validate:

📘 Step-by-step guide to using Phoenix + UQLM: arize.com/docs/phoenix...

⚪️ Spot hallucination-prone responses by surfacing low-confidence outputs ⚪️ Flag uncertain generations for fallback, review, or guardrails ⚪️ Compare prompts & models more rigorously with uncertainty signals ⚪️ Monitor safety + reliability in production by tracking confidence drift

🔍 New Phoenix Evaluation Integration: CVS Health’s UQLM (Uncertainty Quantification for Language Models) metric. UQLM gives you a quantitative measure of LLM uncertainty, so you can:

🔍 New Phoenix Evaluation Integration: CVS Health’s UQLM (Uncertainty Quantification for Language Models) metric. UQLM gives you a quantitative measure of LLM uncertainty, so you can:

Nvidia launched a course on building reliable agentic systems w/ NeMo Agent Toolkit (NAT) It shows how Phoenix helps teams inspect agent reasoning, tool selection, & control flow, making NAT-based agents easier to debug, evaluate, and run in production. Taught by @mcbrayer.bsky.social

Nvidia launched a course on building reliable agentic systems w/ NeMo Agent Toolkit (NAT) It shows how Phoenix helps teams inspect agent reasoning, tool selection, & control flow, making NAT-based agents easier to debug, evaluate, and run in production. Taught by @mcbrayer.bsky.social

🏢 Phoenix now supports on-prem LDAP authentication! For enterprises that can't use cloud SSO, Phoenix integrates directly with your internal directory—Active Directory, OpenLDAP, or any LDAP v3 server. Your data. Your infrastructure. Your identity provider.

In our latest Evals Series webinar, we covered how to evaluate your evaluator. AI development has two loops, Meta-evaluation lives in the inner loop. We also walked through a live demo of this loop in practice, iteratively improving the judge and showing measurable gains at each step.

In the coming weeks we'll be enabling basic telemetry by default in Phoenix over the coming weeks. Our commitment: privacy and security are foundational What we're collecting: Simple, anonymous usage stats Opt out: Set PHOENIX_TELEMETRY_ENABLED=false—no questions asked. github.com/Arize-ai/ph...

Phoenix Evals now supports message-based LLM-as-a-judge prompts— an upgrade that aligns evals with how modern models actually expect instructions. 🧵👇

We just shipped a Span Notes API in Phoenix 12.20. It's a simple API, but it unlocks an important workflow for LLM evaluation: open coding. Here's why this matters.

Check out this walkthrough on bringing observability and evals into LLM workflows, plus a Phoenix demo with helpful context for anyone building agents in TypeScript. Watch the session below 👇