Cursor Cloud Agents add Projects and self-hosted runners, Claude Code 2.1 updates, Codex CLI gains GPT-6.1 Sol (agent tools + practical hook)

What shipped

From 25 September to 2 October 2026, Cursor expanded its cloud agent platform with long-running workflow tools and private deployment options. Claude Code and OpenAI Codex focused on terminal stability, model defaults, and small UX fixes that affect daily agent use.

Cursor

Cursor shipped the largest set of updates with repeated emphasis on cloud agent orchestration. The releases center on Projects for multi-month coordination, self-hosted execution inside private networks, and new bots that review security and track deployments for Teams plans.

• •Automate workflows with event subscriptions Cloud agents now subscribe to PRs and Slack, pin custom modes, spawn subagents in isolated VMs, and accept persistent /goal objectives with non-interrupting steering.
• •Start Cloud Agent projects without external git Agents create Origin repos automatically and add live browser previews plus one-click Vercel publishing for teams that skip third-party SCM setup.
• •Run Cloud Agents on self-hosted machines Tool execution stays inside private networks with dynamic pools, AWS Lambda or Modal sandboxes, and computer use on Linux and Mac.
• •Manage long-running tasks with Projects Coordinators keep months of context while delegating work to thousands of subagents and triggering actions from Slack or schedules.
• •Deploy Rollouts and Security Review bots Teams plans now include bots that monitor deployment health and scan PRs for vulnerabilities, secrets, and auth issues.
• •Add event subscriptions and custom modes Background listeners for PRs and Slack plus pinned skills keep agents responsive without constant user input.
• •Enable cloud agent projects without SCM providers Agents scaffold directly into Origin repos with in-browser previews and direct Vercel pushes.
• •Support self-hosted machines and private pools Dynamic team runners with auto-hibernation let agents run inside company infrastructure.
• •Launch Projects for multi-agent workflows Shared context across local and cloud subagents supports large migrations or full app builds over months.
• •Introduce Rollouts and Security Review bots Rollouts tracks regressions per environment while Security Review flags exploitable bugs in PR context.
• •Add subscriptions, custom modes, and subagent VMs Cursor cloud agents now handle parallel work with isolated VMs and persistent objectives via /goal.
• •Start projects from scratch without external Git Agents initialize Origin repos with port-forward previews and one-click publishing.
• •Add self-hosted machines and team pools Tool calls execute on personal VMs or auto-scaling pools with Linux and Mac computer use.
• •Launch Projects for large-scale coordination Coordinators delegate tasks across cloud subagents while syncing context and reacting to scheduled events.
• •Launch Rollouts and Security Review bots Available on Teams and Enterprise to monitor deploy health and detect injection risks in pull requests.
• •Add event subscriptions and isolated subagents Agents subscribe to external threads, keep custom modes active, and run subagents in dedicated VMs.
• •Add Origin repos with live browser previews Cloud Agents start without external Git and include direct Vercel deployment from the browser.
• •Add self-hosted machines and dynamic scheduling My Machines and Team pools support idle hibernation and multiple sandbox providers.
• •Introduce Projects with persistent context Coordinator agents manage multi-month work and delegate across parallel subagents with shared syncing.
• •Add subscriptions, custom modes, and /goal Cloud agents receive non-interrupting steering and maintain long-lived objectives across sessions.
• •Scaffold projects directly to origin repos Agents bypass third-party SCM and gain browser port-forward previews plus Vercel publishing.
• •Add self-hosted machines and pool scheduling Agents run tool calls inside private networks with custom sandboxes and Mac computer use.
• •Introduce Projects with coordinator agents Long-term work like migrations uses shared context and Slack or schedule triggers.
• •Launch Rollouts and Security Reviewer bots Rollouts reports per-environment health while Security Review enforces codebase rules on PRs.

Claude Code

Claude Code released eighteen incremental updates focused on terminal UX and model defaults. The changes improve subagent handling, add mouse support in lists, and set 1M context as default for newer models on multiple providers.

• •Add subagent backgrounding on Send Now Ctrl+Enter now moves a subagent command to background so the new message processes immediately.
• •Enforce automated verify skill before commits Projects with a verify skill run it automatically before Git commits except for docs or test-only changes.
• •Add mouse support for list pagination Fullscreen lists accept clicks on 'N more' rows with hover states for faster navigation.
• •Display stacked count on permission prompts Multiple queued prompts now show indicators like '2 of 5' to reduce confusion.
• •Add background execution controls to VS Code Active commands gain a Run in background option and surface outputs on the agent map.
• •Enable 1M context window by default Opus 4.7+ and Fable default to 1M on Bedrock, Vertex, and Claude apps without a suffix flag.
• •Add built-in REST gh api for self-hosted runners macOS and Linux sessions without GitHub CLI gain a REST-only gh api command.
• •Support URL prompts from MCP servers The 2025-11-25 protocol now accepts URL prompts for sign-in flows when servers set bareElicitationCapability.
• •Add session and task filter to agents view An n: filter collapses matching sessions and opens the first match on Enter.
• •Introduce Claude Mods for deeper customization Plugins can modify behavior, including a built-in side agent that flags missed items.
• •Update Claude Code to 2.1.285 with WebFetch toggle Desktop launch via --desktop and a flag to disable web fetch arrive with subagent fixes.
• •Update Claude Code to 2.1.286 with UI enhancements Stacked prompts, mouse navigation, and VS Code bookmarks ship alongside session resume fixes.
• •Upgrade to Claude Sonnet 5.5 and track spend limits Version 2.1.284 sets Sonnet 5.5 as default with 1M context and adds /usage spend tracking.
• •Update Claude Code to 2.1.285 with desktop launch CLI plugin commands and API provider restrictions join fixes for artifacts and the VS Code extension.
• •Add prompt auditing and Bedrock Mantle provider /doctor prompt-audit inspects CLAUDE.md files and new gateway headers support load testing.
• •Add Claude Sonnet 5.5 and set default auto mode Interactive sessions now default to auto mode with gateway spend limits visible in /usage.
• •Update Claude Code to 2.1.282 with prose width capping Wide terminals gain a maxProseWidth setting and faster session resume.
• •Update Claude Code to 2.1.283 with prompt auditing Managed settings add model access controls and reduce MCP background notification issues.

OpenAI Codex

OpenAI Codex delivered thirty-one CLI releases that refine terminal behavior and introduce GPT-6.1 Sol as a lower-cost default for long tasks. Updates also improve rendering, session migration, and background server handling.

• •Codex CLI 0.159.0 introduces instant interrupt Users can now interrupt runs mid-response and scroll transcripts during plan reviews with improved Mermaid output.
• •Codex CLI 0.159.1 adds GPT-6.1 Sol default GPT-6.1 Sol becomes the default across bundled, Bedrock Mantle, and Runtime catalogs.
• •Codex CLI 0.159.2 suppresses Windows console flashing Background helper processes no longer flash console windows on Windows.
• •Support GPT-6.1 Sol in Codex and ChatGPT Work The model delivers near-Astra performance for repeated long-running code and document tasks at lower cost.
• •Codex CLI 0.159.3 adds account security reminders Local ChatGPT-authenticated sessions receive optional reminders to complete security setup.
• •Codex CLI 0.160.0 adds task pagination The agent command center gains pagination, X11 primary selection on Linux, and opt-in Guardian review context.
• •Codex CLI 0.158.0 released OpenAI published the 0.158.0 maintenance build with internal fixes.
• •Codex CLI 0.159.0 introduces instant interruption Opt-in instant_interrupt lets users steer responses mid-turn with refreshed welcome screens.
• •Set GPT-6.1 Sol as default in Codex CLI catalogs The model is now preset in bundled and Amazon Bedrock catalogs.
• •Suppress flashing Windows console windows in Codex CLI Sandboxed commands no longer trigger visible console flashes on Windows.
• •Release GPT-6.1 Sol in Codex and ChatGPT Work The release targets complex repeated work across code, apps, and documents at reduced cost.
• •Add account security setup reminders in Codex CLI Eligible local sessions show optional prompts to finish security configuration.
• •Codex CLI 0.158.0: Support MCP OAuth secrets Pre-registered OAuth client secrets and copy-on-select settings arrive with bearer token security for WebSockets.
• •Codex CLI 0.159.0: Add instant interrupt and UI enhancements Transcript scrolling and expanded Mermaid support ship with Windows console suppression.
• •Codex CLI 0.159.1: Default to GPT-6.1 Sol in bundled and Bedrock catalogs The model is set as default for Amazon Bedrock Mantle and Runtime.
• •Codex CLI 0.159.2: Suppress flashing Windows console windows The fix prevents flashes when Codex spawns background processes.
• •Add GPT-6.1 Sol model in Codex and ChatGPT Work Near-Astra performance for long-running tasks is now available via the gpt-6.1-sol identifier.
• •Codex CLI 0.158.0 released with TUI clipboard and security enhancements Copy-on-select, right-click paste, and MCP OAuth support are included.
• •Enhance terminal rendering for lists and math Unicode bullets, checkboxes, and aligned equations improve readability in the TUI.
• •Enable /import command in remote and background sessions Session migration now works across remote and local background-server sessions.
• •Add shortcut to fork conversations Pressing 'f' forks an active conversation from another app while keeping drafts.
• •Add automatic background-server startup Eligible interactive sessions start background servers automatically with recovery options.
• •Enable fullscreen transcripts by default Shift-click extends selections across fullscreen transcript views.
• •Add GPT-6 Sol and Luna with Amazon Bedrock support New models join the catalog with migration prompts for older configurations.
• •Codex CLI 0.157.1 maintenance release The build contains routine internal fixes.
• •Improve terminal rendering for equations and list formatting Unicode support and aligned math notation enhance TUI output.
• •Support /import command in remote and background-server sessions Migration commands now function in both remote and local background sessions.
• •Add keyboard shortcut to fork conversations from other apps The 'f' key preserves unsent drafts when forking sessions.
• •Enable automatic background-server startup for interactive sessions Recovery options appear when server settings are incompatible.
• •Add GPT-6 Sol and Luna models with Amazon Bedrock support The catalog update includes automatic migration prompts for prior models.

What this means for you

For Vibe Builders: Cursor's Projects and self-hosted runners let you run persistent agent teams on your own machines without writing glue code. Claude Code and Codex CLI updates add background execution and instant interrupts so you can steer agents mid-task from the terminal. Start testing Cursor Projects on a small migration this week to see how coordinator agents reduce manual handoffs.

For Non-techies: For day-to-day business work, Cursor now handles long-running tasks like content updates or report generation across multiple agents while keeping data inside your network. Claude Code and Codex CLI changes make terminal agents easier to pause and resume without losing context. Try the new GPT-6.1 Sol default in Codex for cheaper repeated document tasks.

For Developers: Cursor's self-hosted pools and Projects coordinator pattern give production teams a concrete alternative to cloud-only agents with months-long context retention. Claude Code 2.1 series and Codex 0.159-0.160 releases add 1M context defaults, instant interrupts, and MCP OAuth support that you can benchmark against your current runtime. Evaluate the new Security Review bot on your next PR workflow before rolling it to the full team.

What to watch next

Watch for Cursor to expose Projects to individual plans and for Claude Code to ship broader MCP server support. Track whether Codex CLI 0.160 pagination reduces command-center load in daily use.