arize-phoenix: v20.27.0
Arize Phoenix v20.27.0 introduces prompt version timestamps, TypeSafe Jev decision models in Python evaluators, and on-demand PR reviews, alongside security fixes for agents, DSL span filters, and PII redaction.
Why this signal matters
Arize Phoenix has released version 20.27.0. Key features include exposing prompt version creation timestamps in the API, support for TypeSafe Jev decision models in Python classification evaluators, and factory updates for single-verdict PR review grading and on-demand pull request reviews. Significant bug fixes and security-related patches were also deployed: hardening of GitHub token handling and subagent write access in agents, remediation of three span-filter DSL defects surfaced in a security review, and PII redaction from exported logs and spans. Additionally, the release adjusts Anthropic evaluation tool calls (requesting tool calls instead of forcing choices), scopes GraphQL trace error queries by project and time range, enforces graphql-core = 3.3.1, updates arize-phoenix-evals dependencies, and improves client error logging by route.
Actionable summary
Arize Phoenix v20.27.0 adds API exposure for prompt version creation timestamps, supports TypeSafe Jev decision models in Python classification evaluators, and adds PR reviewer factory capabilities. Crucial fixes include hardening GitHub token handling and subagent write access, resolving span-filter DSL security defects, redacting PII from exported logs/spans, and scoping GraphQL trace error subqueries.
- Agent usefulness
- 77/100
- Confidence
- 96%
- Canonical data
- JSON + Markdown
What builders should check
- Upgrade arize-phoenix to v20.27.0, especially if utilizing agent GitHub integrations or span-filter DSLs requiring security patches.
- Verify compatibility with dependency requirement changes, including graphql-core = 3.3.1 and arize-phoenix-evals updates (3.9.4/3.9.5).
- Review exported spans and logs to validate behavior following PII redaction changes.