Codex CLI 0.142.0: Token Budgets, Plugin Recommendations, Indexed Web Search, and Time Awareness
CodexCodex CLI 0.142.0 introduces configurable rollout token budgets that track spending across agent threads and abort turns when limits are reached, giving teams cost control over long-running workflows. The release adds an indexed web-search mode that lets agents perform live searches while restricting direct page access to server-approved URLs. The plugin discovery system gains AI-powered recommendations within eligible turns, and Codex can now query the current time and receive scheduled UTC reminders.
Sources & Mentions
3 external resources covering this update
Codex CLI 0.142.0
This release focuses on cost control, smarter discovery, and a few capability additions for long-running agentic work.
Configurable token budgets
Rollout token budgets track spending across agent threads and abort a turn automatically when a limit is reached. This gives teams predictable, fine-grained cost control over long-running pipelines rather than discovering overruns after the fact.
Indexed web search
A new indexed web-search mode lets agents perform live searches while restricting direct page access to server-approved URLs only. It balances information freshness against access control for compliance-sensitive deployments.
Plugin recommendations
The plugin discovery system now surfaces AI-powered recommendations during eligible turns, so Codex can suggest and install the right plugin mid-session instead of requiring manual catalog browsing. The remote plugin catalog is reorganized into OpenAI Curated, Workspace, and Shared with me sections.
Time awareness and delegation controls
Codex can now query the current UTC time through a clock tool and receive scheduled time reminders, enabling time-sensitive automations. Multi-agent delegation becomes configurable at the thread and turn level (disabled, explicit-request-only, or proactive). The release also lands startup performance improvements through DNS deferral, model cache pre-warming, parallelized skill metadata reads, and reuse of parsed plugin data.