What happened
OpenAI’s release notes for Codex CLI 0.169.0 describe new features, fixes, documentation changes, and maintenance work. In the opt-in terminal workspace, users can now keep multiple side chats in tabs, switch between them with a keyboard or mouse, and reopen saved local side chats after restarting Codex. A side chat can therefore remain available as a separate thread while a developer continues work elsewhere in the terminal interface.
The release also adds search for older tasks from the agent command center. Search can use a task name, a preview of its prompt, or its working directory. Those fields matter because they map to different ways developers tend to recover prior work: remembering a task’s stated goal, recognizing the start of a request, or locating the repository where it ran. The command center had already become part of the product’s multi-agent workflow; this release adds retrieval of prior task context rather than only navigation among current work.
A full-screen transcript option called “Collapse turn details” folds completed activity beneath each prompt while leaving answers visible. Tool output remains searchable. The feature changes presentation rather than removing activity from the transcript: a developer can make a long interaction easier to scan, then use search when the underlying command or tool output becomes relevant. OpenAI also adds a local diagnostic workflow: pressing Ctrl+S in the terminal report editor saves a diagnostic ZIP even when uploads are disabled.
The terminal can now add or edit custom model providers. With custom gateway onboarding enabled, a user can create a provider during sign-in; supported gateways can use device-code authorization, including when connecting over SSH or when no local browser is available. The release documentation also covers app-server clients listing models or authorizing a gateway for a configured provider without changing the default provider. Together, these changes place provider configuration and authentication closer to where developers run the coding agent.
The release extends the opt-in Cloud connection on macOS and Linux. From the terminal, users can send images, choose a model and reasoning effort, and archive or restore Cloud conversations. These controls sit alongside the local-terminal workflow rather than replacing it. Their availability is explicitly tied to the opt-in Cloud connection and to the two named operating systems.
The notes list a set of fixes alongside these workflow changes. codex exec now keeps project configuration and MCP servers disabled in untrusted directories, even when run with --sandbox workspace-write; users can explicitly save project trust with --trust for codex exec or codex review. Other fixes cover Windows sandbox handling for quoted cmd.exe paths and PowerShell sessions with an unreadable parent directory, pasted-image orientation on macOS, interrupted turns, approval cancellations, and waiting on MCP servers that a shared app server did not report. OpenAI published version 0.169.0 on November 13, 2026.
What the archive adds
New Runtime’s earlier coverage of Codex CLI 0.167.0 described its agent command center, cross-backend switching, and local image previews. Version 0.169.0 is a new release on that continuing story: it adds task search, side-chat persistence, and more provider-management workflows rather than restating the earlier feature set.
Trend signal
This release supports agent-ready software exposes capabilities: Codex is making distinct capabilities—searching task history, managing provider configuration, choosing model settings, and handling Cloud conversations—available through its terminal interface. The significance is not a new interoperability standard. It is a product boundary that exposes actionable, permissioned workflows to the person operating the agent.
Why now
The release follows Codex CLI 0.167.0, which introduced an agent command center and backend switching. Version 0.169.0 builds directly on those surfaces: a command center gains historical task search, while provider switching is complemented by provider onboarding and authorization from the terminal. The sequence indicates continued iteration on managing sessions and model access in a terminal-first coding workflow.
Why it matters
Coding agents create a new navigation problem as soon as a developer has more than one task, repository, model route, or ongoing conversation. The agent may generate the code, but people still need to find the right thread, recover its rationale, inspect its tool activity, and decide which model or provider should handle the next step. Side-chat tabs and task search address this operational overhead directly. Searching by working directory is particularly useful when task names are vague but the repository is known.
Custom-provider configuration matters to teams whose model access is mediated through a gateway. Device-code authorization can be practical on a remote machine, in an SSH session, or in environments where opening a browser locally is not convenient. Meanwhile, collapsing completed turn details preserves a review path: the answer stays visible while tool output remains searchable rather than being discarded.
The security behavior is equally relevant to adoption. An agent interface that can load project configuration and MCP servers has an execution boundary to manage, especially when a developer enters an unfamiliar directory. The explicit --trust path keeps that decision visible. Teams should treat the convenience features and the trust controls as one operating model: faster session management should not turn project-level capabilities on implicitly.
What happens next
- Codex will likely add further navigation and management controls around multi-session terminal work. (3 months, 65%). Watch for: additional command-center filters, session metadata, or recovery controls.
- Gateway-oriented provider setup will likely broaden beyond the initial supported gateway paths. (6 months, 55%). Watch for: newly documented provider types or authorization methods.
Limits and uncertainty
OpenAI identifies the terminal workspace and Cloud connection as opt-in, and device-code authorization applies to supported gateways. The release describes capabilities and fixes, not a comparative benchmark of time saved, reliability, adoption, or behavior across every provider, operating system, and team configuration. The workflow effects will depend on local policy, credentials, gateway support, and project layout.
The security item specifies how codex exec treats project configuration and MCP servers in untrusted directories and how a user may save trust. It should not be read as an independent security assessment of the CLI, a gateway, or an MCP server. Nor has the practical value of the side-chat and search changes been independently measured in this Field Note.
Skill to build now
Map one recurring Codex task to a recoverable terminal workflow this week. Give the task a distinctive name, work in a consistent repository directory, and verify that it can be found from the agent command center using both identifiers. If your team uses a gateway, test provider onboarding and device-code authorization in a non-production environment first. Keep untrusted-project work separate and require an explicit review before using --trust.
Reader takeaway
Codex CLI 0.169.0 is an interface-and-operations release, not a new model launch. Its useful additions are persistent parallel side chats, searchable task history, and terminal-based provider setup. If Codex is part of your daily development loop, test whether those controls reduce the cost of resuming work while preserving explicit project trust boundaries.
Open question
Will task search and side-chat persistence become sufficient for teams to manage a growing queue of agent work, or will Codex need stronger shared review, ownership, and handoff controls?