Safety ResearchAutomated release watch

Promptfoo 0.124.0

Promptfoo 0.124.0 removes the hosted ChatKit provider and transitions several integrations (Langfuse, Slack, WatsonX, OpenAI Agents, browser SDKs, and local Transformers) to opt-in explicit installs, alongside expanding model support and workspace isolation for coding agents.

Human read

Why this signal matters

Promptfoo version 0.124.0 includes significant breaking changes focused on dependency minimization and provider modularization. The hosted ChatKit provider has been removed entirely. Multiple third-party and provider dependencies—including OpenAI Agents, browser SDKs, WatsonX SDKs, Langfuse SDK, Slack SDK, Codex Security SDK, and local Transformers—have been transitioned to opt-in or explicit installs. On the feature side, the update introduces isolated workspaces for coding agents, the ability to resume paused Claude server-tool turns, support for new models (Claude Opus/Sonnet 5.5, GPT-6 Sol/Luna, GPT-6.1 Sol/ultrafast, Grok 4.7), and TypeSafe Jev provider integration. It also patches several assertion grading workflows, dependency vulnerabilities, and CLI behaviors.

Agent parse

Actionable summary

Promptfoo 0.124.0 introduces breaking modularization changes by requiring explicit/opt-in installations for OpenAI Agents, browser SDKs, WatsonX, Langfuse, Slack, Codex Security, and local Transformers, while completely removing the hosted ChatKit provider. It also adds workspace isolation for coding agents, server-tool resumption for Claude, and support for newer model versions.

Agent usefulness
94/100
Confidence
96%
Canonical data
JSON + Markdown
Next actions

What builders should check

  • Remove any usage of the deprecated hosted ChatKit provider.
  • Explicitly install packages for OpenAI Agents or browser SDKs if utilized in test suites.
  • Configure opt-in installations for WatsonX, Langfuse, Slack, Codex Security, or local Transformers if your evaluations depend on them.
Classification

Tags and routing

evaluationred-teamtesting
Related signals

Continue the thread