LoginSign Up
← Back to all sparks
A

Arize Phoenix

AI-ML-PLATFORMS
Velocity5.0

AI observability and evaluation platform for LLM applications

phoenix.arize.com ↗

Phoenix evals pins out a compromised LiteLLM release and the emSign root.

llm observabilityevalstracingsupply-chain securityopen source
◆Current state
Arize Phoenix is shipping near-daily releases across its server, Python evals and TypeScript packages. The past week was mostly fixes and dependency hygiene: evals 3.9.3 excluded the compromised LiteLLM 1.107.0 and required a certifi build that distrusts the emSign root, and later patches refreshed playground presets and evals kwargs handling. Server 20.25 and 20.26 mainly carried the team's internal issue-to-PR agent tooling, plus a service.version attribute on Phoenix's own OpenTelemetry resource.
◆Where it's heading
Phoenix keeps widening tracing coverage through integration pages (Pipecat, Microsoft Agent Framework) and its TypeScript evals, which now run classifiers on AI SDK decision models. The core product sees small, frequent fixes rather than large features right now, and supply-chain pinning has become part of the release routine.
◆Prediction
The TypeScript decision-model support is pinned to experimental AI SDK versions, so expect follow-up patches as that API changes.

◆Recent moves

  1. 1d ago

    Phoenix MCP 4.3.18: client dependency bump

    The Phoenix MCP server only picks up the client 7.16.3 schema fix as a dependency bump.

    View source ↗
  2. 1d ago

    Phoenix CLI 1.18.10: client dependency bump

    The Phoenix CLI takes the same client 7.16.3 dependency bump, with no change of its own.

    View source ↗
  3. 1d ago

    Phoenix evals 2.7.1 keeps custom evaluator metadata

    The TypeScript evals package now keeps metadata returned by custom evaluators wrapped with createEvaluator. A small correctness fix in the run of frequent evals patches.

    View source ↗
  4. 1d ago

    Phoenix client 7.16.3 keeps schema keywords beside anyOf

    The client's toSDK conversion now keeps description, default and other keywords next to anyOf in prompt tool and response-format schemas. A schema-fidelity fix for teams exporting Phoenix prompts to provider SDKs.

    View source ↗
  5. 1d ago

    Evals 3.9.5: PrecisionRecallFScore docstring fixes

    A docs-only evals patch fixing docstring examples for PrecisionRecallFScore. More of the small, frequent fixes that fill Phoenix's current releases.

    View source ↗
  6. 3d ago

    Phoenix 20.26.1: MiniMax presets, Agent Framework tracing docs

    20.26.1 refreshes MiniMax presets in the playground and adds a Microsoft Agent Framework tracing page. It is a docs-and-presets patch.

    View source ↗