OpenAI released version 0.169.0 of its terminal agent Codex CLI (rust-v0.169.0) on November 13, 2026. As documented in the official release notes, the update introduces two major shifts: enforced project trust boundaries to neutralize malicious repository configs, and a multi-tab side-chat interface in the terminal workspace.
1. Disabling Project Configs and MCP in Untrusted Directories
Cloning public repositories to review or debug code locally introduces an attack surface if the codebase includes a compromised configuration file (.codex/config.toml) or malicious Model Context Protocol (MCP) server declarations.
Starting with v0.169.0, codex exec and codex review enforce strict restrictions whenever the current working directory has not been explicitly marked as trusted:
- Project-Level Settings Ignored: Repository-specific configuration overrides are suppressed, falling back solely to the user's global configuration.
- Automatic Disabling of Local MCP Servers: Any MCP servers or tool definitions packaged inside the repository remain inactive, preventing untrusted code from triggering unexpected network calls or host shell executions.
- Explicit Opt-in: Activating project-specific tooling requires passing the
--trustflag explicitly or confirming via the interactive trust prompt.
2. Terminal Side-Chat Tabs and Local Session Resumption
The terminal interface workspace (features.tui_workspace) has been upgraded to support multiple concurrent side-chat tabs (PR #59739):
- Tab Navigation: Developers can switch between different conversational threads using keyboard shortcuts or mouse clicks, separating distinct debugging tasks without context pollution.
- Persistence Across Restarts: Local side-chat sessions are preserved on disk and can be resumed after closing and relaunching the CLI.
3. Past Task Search
The agent command center now includes an indexed task search. Previous runs can be retrieved by task name, initial prompt preview, or the working directory where the execution occurred, reducing the need to manually dig through historical terminal buffers.
4. Custom Model Providers and Device-Code Auth
Configuration menus for custom LLM gateways and private enterprise endpoints can now be configured directly within the terminal. Additionally, for headless servers and SSH environments where browser-based OAuth is impractical, a device-code authorization flow has been introduced.
Operational Comparison
| Feature | Prior Versions (v0.168.x) | v0.169.0 |
|---|---|---|
| Untrusted Folder MCP | Loaded automatically from repo | Disabled by default (requires --trust) |
| Project Settings | Read without prompt | Isolated to global defaults until trusted |
| Task History | Manual log review | Indexed search by prompt, name, directory |
| Terminal UI | Single-stream workspace | Multi-tab side chats with state persistence |
Source: OpenAI Codex CLI v0.169.0 Release.