SELF-HOSTED MACHINES -- CURSOR AGENTS CAN NOW RUN ENTIRELY INSIDE YOUR OWN NETWORK (2026-09-02, vendor-primary): Cursor shipped **self-hosted machines**, which 'let you keep tool execution entirely in your own network. Your codebase, build outputs, and secrets all stay on internal machines running in your infrastructure, while the agent handles tool calls locally.' **THE TIMING IS WORTH NOTING GIVEN EVERYTHING ELSE ON THIS PAGE.** Cursor is now a SpaceX subsidiary and is being cut off by OpenAI, and the single most common enterprise objection to both facts is where the code goes. This ships a real answer to that objection five days after the OpenAI wind-down notice -- it does not restore the model lineup, but it removes the data-residency argument against staying. **WHAT IS ACTUALLY IN IT:** **Dynamic pool scheduling** -- 'My Machines' connects one laptop or VM for personal use, while **team pools** are named queues of workers that grow as requests arrive and shrink as workers disconnect. Pools are **not tied to one repository** ('name the pool, and any available worker can claim the request') and can **hibernate idle machines**, restoring them within a reconnect window, so you are not paying to keep capacity warm between prompts. **Run on your own sandboxes:** cloud agents can execute on **AWS Lambda, Coder, Cloudflare, Daytona, Modal, Namespace, Vercel and E2B**. **Computer use on Linux and Mac** for self-hosted workers -- with the right desktop packages an agent can click, type, screenshot and drive a browser, and you can watch its desktop or take control from Cursor. **The practical read: this is an enterprise-retention feature, and a well-judged one -- but it addresses trust and control, not the November 12 loss of OpenAI models, which remains the live migration question for anyone whose Cursor workflow depends on GPT specifically.**
2026-09-02•Cursor (cursor.com/changelog/self-hosted-machines, on-page 'Sep 2, 2026') -- fetched 2026-09-05 via curl with browser UA
OPENAI IS CUTTING CURSOR OFF -- CONTRACT WIND-DOWN NOTIFIED 2026-08-28, PROPOSED SHUTOFF 2026-11-12 (vendor-primary, OpenAI): **this is the largest risk to Cursor's model lineup since the product launched, and it is a direct consequence of the SpaceX acquisition that closed on 8/14.** OpenAI states it plainly: '**Today, we notified SpaceX that we intend to wind down our contract providing OpenAI models to Cursor, with a proposed shutoff date of November 12, 2026.**' **Two things about the mechanism are worth understanding before you plan around it.** First, **this is the maximum notice OpenAI's contract allows, not a negotiated runway** -- OpenAI says it is 'giving the maximum notice provided by our contract' specifically 'to maximize the time that developers can retain access to our models through Cursor', and that its custom agreement 'gives us a limited time window to cancel it after a change of control'. So the date is a contractual ceiling, and there is no indication of an extension path. Second, and more immediately: **OpenAI says it is 'not providing future models to Cursor' from now**, separate from the November cutoff. **That means the practical degradation starts immediately, not on November 12** -- Cursor's OpenAI lineup is frozen at whatever shipped before 8/28, so any GPT model OpenAI releases between now and the shutoff will simply not appear in Cursor. **OPENAI'S STATED REASON IS CONTRACT-COMPLIANCE HISTORY, NOT PRODUCT OR PRICING:** it says it 'cannot be confident that SpaceX will use our technology within our terms of service, based on our experience with Elon Musk's companies violating contracts', citing Twitter breaking contract terms after Musk's acquisition and Musk's admission under oath earlier this year that xAI had violated OpenAI's terms of service. OpenAI also ties the decision to its **upcoming Astra model** and 'a new level of accountability to ensure [it] is being used in accordance with our terms'. **WHAT THIS MEANS IF YOU PAY FOR CURSOR:** model neutrality has been Cursor's core buying argument, and it is being dismantled from both ends -- the SpaceX/xAI side is pushing Grok deeper in (Grok 4.6 day-one, Grok Bot now bundled to Pro), while OpenAI is exiting. **If your Cursor workflow depends on GPT models specifically, treat November 12 as a real migration deadline and note that new OpenAI models are already off the table.** Anthropic and Google models are unaffected by this announcement. OpenAI adds that it has 'enormous respect' for the Cursor team after nearly four years and is 'ready to go above and beyond' to support affected developers, but commits to no specific remedy. **Cursor and SpaceX had published no response on cursor.com as of this check.**
2026-08-28•OpenAI (openai.com/index/our-decision-on-cursor-following-its-acquisition-by-spacex/, publicationDateText 'August 28, 2026') -- fetched 2026-08-31 via curl with browser UA; cursor.com/blog and cursor.com/changelog both enumerated the same session and carry no response post
CLOUD AGENTS NO LONGER NEED A REPO -- 'START FROM SCRATCH' SHIPPED (2026-08-27, vendor changelog): Cursor removed the connected-SCM requirement for Cloud Agents. Previously you needed a linked GitHub or other third-party SCM provider before an agent could start; now you '**Select Start from scratch in the repo picker, then prompt the agent**', and Cursor creates a **Cursor Origin** repo behind the scenes. When you like what the agent built, **Create repo** saves it into a fully scaffolded Origin repo with a custom or suggested name and **private or internal** visibility. Two capabilities shipped alongside it: Cursor now **port-forwards the cloud agent's live environment straight to your browser** for preview and design-mode work, and you can **connect a Vercel account and publish to a live URL** from the same page. **The strategic read is vertical integration, not convenience.** Cursor Origin is Cursor's own code hosting (the 8/18 'Git at any scale' post described its architecture), and this change makes Origin the **default destination for greenfield work** rather than an opt-in migration -- a new project now starts and stays inside Cursor unless you deliberately move it out. **Caveat to note before relying on the publish flow: a Vercel account is required**, so 'publish' is a Vercel integration, not a Cursor-hosted deployment.
2026-08-27•Cursor (cursor.com/changelog/start-from-scratch, on-page 'Aug 27, 2026 · Changelog') -- fetched 2026-08-31 via curl with browser UA
GROK BOT NOW INCLUDED WITH EVERY PAID CURSOR TIER, INCLUDING PRO (2026-08-26, vendor-primary): xAI extended Grok Bot to **Cursor Pro, Cursor Pro+, Cursor Ultra and all Cursor Teams plans (Standard and Premium)** -- previously it reached only Cursor Ultra and Teams Premium. **For this page the material fact is that Grok Bot is now a Pro-tier benefit**, so an ordinary paying Cursor subscriber gets xAI's agent product bundled without an upgrade. **And the usage is additive, not shared: 'Grok Bot comes with its own usage, separate from your Grok and Cursor plans, so anything you hand off to a Bot won't count against your existing usage.'** **This is the clearest product-level consequence of the SpaceX/xAI/Cursor consolidation so far** -- the acquisition closed 2026-08-14, and twelve days later xAI's agent product is bundled down to Cursor's entry paid tier. Announced on xAI's newsroom, not Cursor's; Cursor's own blog carries nothing on it, which is exactly why both newsrooms have to be enumerated separately.
2026-08-26•xAI (x.ai/news/grok-bot-more-plans, on-page 'Aug 26, 2026') -- fetched 2026-08-28 via curl with browser UA; cursor.com/blog enumerated same session and carries no corresponding post
THE SPACEX ACQUISITION IS CLOSED -- CURSOR IS NOW PART OF SPACEX (2026-08-14, vendor-primary): the ownership question this page has tracked since April is settled. Cursor's own post is unambiguous in its first line -- '**Cursor has officially been acquired by SpaceX. This completes the acquisition process that started in April, when we announced our partnership with SpaceXAI to accelerate our model training efforts.**' **THIS SUPERSEDES THE 'OPTION TO ACQUIRE' FRAMING FURTHER DOWN THIS PAGE.** The 2026-04-21 disclosure was an *option* -- $60B all-stock, or $10B in lieu for joint Colossus compute -- with a closing window pegged to SpaceX's June IPO and a 6/29 note that it was 'expected to close in Q3 2026'. It exercised, and it closed inside that window. Anyone still reading this page as 'Cursor might be bought' should stop; the correct present tense is **Cursor is a SpaceX subsidiary**, with Anysphere remaining as the legal entity name in Cursor's own footer. **WHAT CURSOR SAYS CHANGES, IN ITS WORDS:** '**We will have access to the largest fleet of GPUs in the world, giving us the compute to build stronger models that are also more economical to run. This means we can provide customers with more capable models at lower cost.**' It names **Grok 4.6, released 2026-08-12, as 'an early look at what we can now build together'** -- consistent with the 8/11-8/12 entry above, where Grok 4.6 held Grok 4.5's $2/$6 pricing across a capability bump. **WHAT TO ACTUALLY WATCH, BECAUSE THE POST DOES NOT SAY:** no pricing change, no plan change, no roadmap date and no statement on model neutrality accompanies this. Cursor Router still routes across Anthropic, OpenAI and Google models, so the model-agnostic editor is intact **today**. **THE BUYER-SIDE RISK IS UNCHANGED IN KIND AND HIGHER IN DEGREE:** the concern recorded on this page since April -- that a subscription increasingly bundled with one vendor's models is a different product from the neutral editor Cursor was -- no longer depends on a deal closing, because the deal has closed. If your multi-year commitment to Cursor assumes vendor neutrality, that assumption is now a commercial preference of SpaceX rather than a structural property of the company. **A promised cost benefit is not a shipped price cut** -- treat 'more capable models at lower cost' as intent until a rate card moves
2026-08-14•Cursor (cursor.com/blog/joining-spacex, dated Aug 14, 2026, fetched 2026-08-20 via curl)
CURSOR ORIGIN GETS ITS ARCHITECTURE POST -- BUT STILL NO GA DATE AND STILL NO PRICING (2026-08-18): Cursor published a 27-minute engineering deep-dive on hosting Git at scale by Vicent Marti, which doubles as the fullest public description yet of **Origin**, the agent-first Git forge announced 2026-06-16 and still waitlist-only. **THE TECHNICAL CLAIM:** Cursor argues that Git's packfile-based storage -- 'large binary files that must exist on a filesystem for Git to access them' -- is the root scalability and availability limit for anyone hosting repositories, and that Origin is built to avoid it. Cursor's framing of why this matters now is the agent workload itself: '**Agents have fundamentally changed the way we work with software, and in many ways they've made this situation worse. More code, more PRs, more CI runs.**' Positioning is a migration story -- '**the smoothest possible off-ramp**' from an existing host -- and Cursor states '**Origin is not an experiment**'. **WHAT IS DELIBERATELY NOT IN THE POST, AND WHY WE ARE NOT SHIPPING IT AS A LAUNCH:** no general-availability date (the 6/16 target of 'fall 2026' is neither restated nor moved), **no pricing**, no migration tooling announcement, no SLA and no open sign-up -- it remains waitlist-only. Filed by Cursor under 'research', not 'product'. Read it as evidence Origin is seriously staffed and architecturally distinct, **not** as evidence you can adopt it yet
2026-08-18•Cursor (cursor.com/blog/git-at-any-scale, dated Aug 18, 2026, filed under research, fetched 2026-08-20 via curl)
Issues here are sourced from our editorial sweeps, not real-time telemetry. Newer issues may exist.