--- title: "OpenAI Codex 0.142 Adds Rollout Token Budgets, Multi-Agent Delegation, Indexed Web Search, and a Reorganized Plugin Marketplace" description: "The OpenAI Codex 0.142 line shipped between June 22 and June 26, 2026 (rust-v0.142.0 through rust-v0.142.3 on github.com/openai/codex). The cycle turns the agent into something a team can govern: configurable rollout token budgets that track usage across threads and abort turns when exhausted, multi-agent delegation configurable as disabled, explicit-request-only, or proactive, an indexed web-search mode that permits live search while restricting page access to server-approved URLs, MCP tools that use tool search by default, a /plugins marketplace reorganized into OpenAI Curated, Workspace, and Shared with me sections, system proxy (PAC, WPAD, bypass) support on Windows and macOS for authentication, and a long list of remote-executor reliability and security fixes. It builds directly on the 0.141 Noise-encrypted relay and cross-OS PathUri layer from June 18." date: 2026-06-27 image: "/images/heroes/2026-06-27--openai-codex-0-142-token-budgets-multi-agent-web-search.png" author: lschvn tags: ["ai", "tooling"] tldr: - "The Codex 0.142 line opened on June 22 ([rust-v0.142.0](https://github.com/openai/codex/releases/tag/rust-v0.142.0)) and reached stable at [0.142.3 on June 26](https://github.com/openai/codex/releases/tag/rust-v0.142.3), a week after the [0.141 Noise relay and cross-OS exec release](/articles/2026-06-18--openai-codex-0-141-noise-relay-cross-platform-exec). The cycle's organizing theme is agent governance and orchestration rather than a single headline security primitive: configurable rollout token budgets track usage across agent threads, remind you of remaining budget, and abort turns when the budget is exhausted ([#28746](https://github.com/openai/codex/pull/28746), [#28494](https://github.com/openai/codex/pull/28494), [#28707](https://github.com/openai/codex/pull/28707), [#29423](https://github.com/openai/codex/pull/29423))." - "Two new orchestration capabilities land on top. Multi-agent delegation is now configurable per thread and per turn as disabled, explicit-request-only, or proactive ([#28685](https://github.com/openai/codex/pull/28685), [#28792](https://github.com/openai/codex/pull/28792), [#29324](https://github.com/openai/codex/pull/29324)). An indexed web-search mode permits live searches while restricting direct page access to server-approved URLs ([#28489](https://github.com/openai/codex/pull/28489)). MCP tools now use tool search by default when the provider supports it ([#29486](https://github.com/openai/codex/pull/29486)), which matters for installs with dozens of MCP servers where the tool list outgrew the context window." - "The rest is plumbing that a team running Codex at scale will feel: the `/plugins` view reorganizes into OpenAI Curated, Workspace, and Shared with me sections and can recommend plugins mid-turn ([#26703](https://github.com/openai/codex/pull/26703), [#28399](https://github.com/openai/codex/pull/28399), [#28400](https://github.com/openai/codex/pull/28400)); authentication honors Windows and macOS system proxy, PAC, and WPAD settings ([#26708](https://github.com/openai/codex/pull/26708), [#26709](https://github.com/openai/codex/pull/26709)); the Linux TUI survives a Ctrl+Z suspend and resume ([#28342](https://github.com/openai/codex/pull/28342)); and exec-server and stdio MCP sessions now survive transient disconnects with signed-URL refresh and retry-safe stdin writes ([#28512](https://github.com/openai/codex/pull/28512), [#28374](https://github.com/openai/codex/pull/28374), [#28546](https://github.com/openai/codex/pull/28546), [#28895](https://github.com/openai/codex/pull/28895)). 0.142.2 also refreshes the bundled OpenSSL to 3.6.3 and esbuild to 0.28.1." faq: - question: "What is new in OpenAI Codex 0.142?" answer: "The 0.142 line (0.142.0 on June 22 through 0.142.3 stable on June 26, 2026) is a governance and orchestration release. It adds configurable rollout token budgets that track usage across agent threads, remind you of remaining budget, and abort turns when exhausted; multi-agent delegation configurable per thread and turn as disabled, explicit-request-only, or proactive; an indexed web-search mode that allows live search while restricting direct page access to server-approved URLs; MCP tools that use tool search by default when supported; and a `/plugins` view reorganized into OpenAI Curated, Workspace, and Shared with me sections. It also adds Windows and macOS system proxy (PAC, WPAD, bypass) support for authentication, scheduled UTC time reminders, and a batch of remote-executor reliability and security fixes." - question: "What are rollout token budgets, and why do they matter?" answer: "A rollout token budget is a configurable cap on how many tokens an agent run (a rollout) can consume across threads. Codex 0.142 tracks usage against the budget as the run progresses, emits a remaining-budget reminder at a configurable threshold, and aborts the turn cleanly when the budget is exhausted rather than letting a runaway task burn through cost. For teams running Codex against paid models, this is the first built-in cost-control primitive at the agent layer: you can bound a background task or a delegation without an external wrapper script. The feature is implemented across four PRs (#28746, #28494, #28707, #29423) and includes turn-scoped context contributions so the budget reflects what the model actually sampled." - question: "What is the indexed web-search mode?" answer: "PR #28489 adds an indexed web-search mode that lets Codex run live web searches while restricting direct page retrieval to server-approved URLs. The model can gather fresh information from the web, but cannot follow arbitrary links off the approved set, which is the shape a team wants when it trusts search aggregation but not unrestricted browser access. Codex distinguishes cached and live web access terminology explicitly in the config, and the mode is separate from the existing sandboxed filesystem and network approval scopes." - question: "How does the new multi-agent delegation work?" answer: "Multi-agent delegation lets a parent Codex thread hand work to a subagent. In 0.142, app-server clients can configure the behavior at both the thread and the turn level as one of three modes: disabled, explicit-request-only (the parent must ask for a delegation), or proactive (the parent may delegate on its own initiative when it judges a subagent fits). A related fix surfaces terminal subagent errors back to the parent agent instead of letting a failed delegation appear as an empty successful completion (#28375), so a parent that delegated to a child that crashed no longer silently continues as if the work succeeded." - question: "Do MCP tools and plugins change for existing users?" answer: "Two changes are visible. MCP tools now use tool search by default when the MCP provider supports it (#29486), so an install with many MCP servers no longer dumps every tool into the model's context up front; the model searches for the relevant tool instead. This preserves compatibility with older models and providers that do not support tool search. The `/plugins` view reorganizes remote plugins into OpenAI Curated, Workspace, and Shared with me sections and can recommend and install a relevant plugin mid-turn based on what the agent is doing (#26703, #28399, #28400, #27704, #28403). 0.142.2 adds dedicated dark-mode logos for plugins (#29488) and curated featured-plugin rankings from remote catalogs (#29485)." - question: "What changes for developers behind a corporate proxy?" answer: "0.142.1 adds opt-in Windows system proxy support for authentication, including PAC, WPAD, static proxy configuration, and bypass rules (#26708). 0.142.2 adds the macOS equivalent: authentication clients honor system proxy, PAC, and WPAD settings when respect_system_proxy is enabled (#26709). Both cover the case where a developer is on a corporate network with an explicit proxy or an auto-config script, which previously broke Codex's auth handshake. 0.142.0 also honors startup custom CA bundles with the managed MITM path (#29014), relevant on networks that intercept TLS." - question: "Is this a breaking release, and how do I upgrade?" answer: "No public API is removed; the changes are additive features, new config surfaces, and fixes. Upgrade with the standard Codex update path, which pulls the latest stable. If you pinned 0.141 for the Noise relay behavior, 0.142 keeps that transport and builds on it, so there is no reason to stay behind. The only behavioral shift to plan for is MCP tool search being on by default: if you depend on every MCP tool being present in the context regardless of relevance, verify that your provider advertises tool search correctly, since the fallback path keeps the old behavior." --- The OpenAI Codex 0.142 line shipped across the week of June 22 to June 26, 2026: [0.142.0](https://github.com/openai/codex/releases/tag/rust-v0.142.0) on June 22, [0.142.1](https://github.com/openai/codex/releases/tag/rust-v0.142.1) and [0.142.2](https://github.com/openai/codex/releases/tag/rust-v0.142.2) on June 25, and [0.142.3](https://github.com/openai/codex/releases/tag/rust-v0.142.3) on June 26. It lands one week after the [0.141 Noise relay and cross-OS exec release](/articles/2026-06-18--openai-codex-0-141-noise-relay-cross-platform-exec), and the contrast between the two is telling. Where 0.141 was a security release whose headline was a default end-to-end encrypted transport between the orchestrator and remote executors, 0.142 is a governance and orchestration release. The question it answers is no longer "is the wire safe" but "can a team run many of these agents and stay in control of cost, delegation, and tool sprawl." ## Rollout token budgets: cost control at the agent layer The most consequential feature is the rollout token budget, implemented across [#28746](https://github.com/openai/codex/pull/28746), [#28494](https://github.com/openai/codex/pull/28494), [#28707](https://github.com/openai/codex/pull/28707), and [#29423](https://github.com/openai/codex/pull/29423). A rollout is a single agent run. The budget is a configurable token cap that Codex tracks across threads as the rollout progresses. At a configurable threshold it emits a remaining-budget reminder, and when the budget is exhausted it aborts the turn cleanly instead of letting a runaway task keep sampling. Turn-scoped context contributions ([#28911](https://github.com/openai/codex/pull/28911)) keep the accounting honest, and the budget window context is wrapped separately ([#29494](https://github.com/openai/codex/pull/29494)) so the model sees what it has left. This is the first built-in cost-control primitive at the Codex agent layer. Before 0.142, bounding what a background task or a delegation could spend meant an external wrapper script or a hard model-side limit. Now a team can hand an agent a budget and trust that a stuck loop will stop on its own. For anyone running Codex against paid models, or running it unattended inside a CI or a [multi-agent orchestration setup](/articles/2026-06-22--sakana-fugu-multi-agent-orchestration-frontier), that is the difference between an agent that is safe to leave running and one that is not. ## Multi-agent delegation and indexed web search Two orchestration capabilities build on the new budget. Multi-agent delegation lets a parent thread hand work to a subagent, and 0.142 makes the behavior configurable at both the thread and the turn level as disabled, explicit-request-only, or proactive ([#28685](https://github.com/openai/codex/pull/28685), [#28792](https://github.com/openai/codex/pull/28792), [#29324](https://github.com/openai/codex/pull/29324)). The "explicit-request-only" mode is the conservative default a team wants when it is evaluating delegation: the parent has to ask before spawning a child. Proactive mode hands the parent the discretion to delegate when it judges a subagent fits the work. A companion fix ([#28375](https://github.com/openai/codex/pull/28375)) surfaces terminal subagent errors back to the parent, so a child that crashed no longer reports as an empty successful completion and leaves the parent continuing on a false premise. The indexed web-search mode ([#28489](https://github.com/openai/codex/pull/28489)) gives the agent live web access without unrestricted browsing. Codex can run live searches, but direct page retrieval is restricted to server-approved URLs. The model gathers fresh information from the web but cannot follow arbitrary links off the approved set. Codex distinguishes cached and live web-access terminology in the config ([#29095](https://github.com/openai/codex/pull/29095)), and the mode sits alongside the existing filesystem and network approval scopes rather than replacing them. The shape is deliberate: trust the search aggregation, do not trust the open browser. ## MCP tool search by default, and a reorganized plugin marketplace MCP tooling gets two changes that matter for installs that have grown past a handful of servers. MCP tools now use tool search by default when the provider supports it ([#29486](https://github.com/openai/codex/pull/29486)). Instead of dumping every tool from every connected MCP server into the model's context up front, Codex searches for the relevant tool on demand. That preserves compatibility with older models and providers that do not advertise tool search, but for a mature install it is the difference between a tool list that fits the context window and one that does not. The `/plugins` view reorganizes remote plugins into three sections, OpenAI Curated, Workspace, and Shared with me ([#26703](https://github.com/openai/codex/pull/26703)), and a turn can now recommend and install a relevant plugin mid-run ([#28399](https://github.com/openai/codex/pull/28399), [#28400](https://github.com/openai/codex/pull/28400), [#27704](https://github.com/openai/codex/pull/27704), [#28403](https://github.com/openai/codex/pull/28403)). 0.142.2 adds dedicated dark-mode logos for plugins ([#29488](https://github.com/openai/codex/pull/29488)) and curated featured-plugin rankings from remote catalogs ([#29485](https://github.com/openai/codex/pull/29485)). The reorganization turns the plugin browser from a flat list into something a team can actually navigate, and the in-turn recommendation closes the loop between "I have a plugin for that" and the agent actually using it. ## System proxy, time awareness, and remote-executor reliability The remaining work is the kind that does not make a headline but decides whether Codex runs on a given network. 0.142.1 adds opt-in Windows system proxy support for authentication, including PAC, WPAD, static proxies, and bypass rules ([#26708](https://github.com/openai/codex/pull/26708)). 0.142.2 adds the macOS equivalent: authentication clients honor system proxy, PAC, and WPAD when `respect_system_proxy` is enabled ([#26709](https://github.com/openai/codex/pull/26709)). 0.142.0 also honors startup custom CA bundles with the managed MITM path ([#29014](https://github.com/openai/codex/pull/29014)), which matters on corporate networks that intercept TLS. Together these cover the proxy-and-CA matrix that previously broke the auth handshake for a large class of enterprise users. Codex also gains real time awareness: scheduled UTC time reminders and the ability to query the current time directly, including through client-provided app-server clocks ([#28822](https://github.com/openai/codex/pull/28822), [#28824](https://github.com/openai/codex/pull/28824), [#28835](https://github.com/openai/codex/pull/28835), [#29011](https://github.com/openai/codex/pull/29011)). The `/usage` command can now show and redeem earned usage-limit reset credits ([#28154](https://github.com/openai/codex/pull/28154), [#28793](https://github.com/openai/codex/pull/28793)). The reliability fixes target the remote-executor path that 0.141 made the default. Exec-server processes and stdio MCP sessions now survive transient disconnects, with signed-URL refresh on reconnect and retry-safe stdin writes ([#28512](https://github.com/openai/codex/pull/28512), [#28374](https://github.com/openai/codex/pull/28374), [#28546](https://github.com/openai/codex/pull/28546), [#28895](https://github.com/openai/codex/pull/28895)). The Linux TUI no longer breaks after a `Ctrl+Z` suspend and `fg` resume ([#28342](https://github.com/openai/codex/pull/28342)). Remote environments now preserve executor-native paths, shells, `AGENTS.md` discovery, and sandbox behavior across operating systems, which is the payoff from the `PathUri` layer 0.141 introduced. ## Security hardening and what to watch 0.142.2 hardens a few spots worth noting. PowerShell commands containing executable AST regions the safety classifier cannot inspect now require approval ([#24092](https://github.com/openai/codex/pull/24092)). Remote HTTP(S) image inputs return clear, model-visible validation errors instead of failing silently, while inline data URLs and local images keep working ([#29417](https://github.com/openai/codex/pull/29417), [#29419](https://github.com/openai/codex/pull/29419)). Expired Amazon Bedrock credentials now produce actionable recovery guidance instead of a generic authorization error ([#28992](https://github.com/openai/codex/pull/28992)). The release also refreshes the bundled OpenSSL to 3.6.3 ([#29487](https://github.com/openai/codex/pull/29487)) and esbuild to 0.28.1, the latter being the same version that shipped the [Deno RCE and Windows path-traversal fixes](/articles/2026-06-14--esbuild-0-28-1-deno-rce-windows-path-traversal) we covered earlier this month. Read alongside the [0.141 Noise relay release](/articles/2026-06-18--openai-codex-0-141-noise-relay-cross-platform-exec) and the broader move of runtimes into [Anthropic's AI coding infrastructure](/articles/2026-04-19--bun-joins-anthropic-ai-coding-infrastructure), the 0.142 cycle is the moment Codex stops being a single-agent coding tool and starts being an agent platform you can govern: budgets to bound cost, delegation modes to control who spawns whom, a web-search mode that trades reach for safety, and tool search that scales to a real MCP estate. The full changelog between 0.141 and 0.142.0 is on the [compare view](https://github.com/openai/codex/compare/rust-v0.141.0...rust-v0.142.0), and the 0.143.0 alpha line is already on its twenty-sixth pre-release at time of writing, which is the signal that the cadence is not slowing down.