Think inside your AI world.
How do I get notified when an API I use ships a breaking change?
An upstream API or library ships a breaking change and it is announced the moment it is cut: a release note, a changelog entry, a status-page post. None of it is in your repository. What you want to hear about is narrow, a major version or a security fix, on the few dependencies you actually rely on, and that narrowness is a rule.
The construction is the public release feeds plus your own filter. GitHub releases are keyless, with an honest rate cap; status pages and changelogs are public too. Your rule is “major or security, on my five repositories.” Through Unl a matching release arrives beside that rule when you are about to touch the integration, stamped with when it was cut. This is the neighbour of the filesystem boundary: last night’s upstream release is not in your working tree, so no search of the machine can fetch it.
The person
A developer who depends on a handful of upstream APIs and libraries and does not want every patch release across every dependency landing at once. They want to hear about the changes that would actually break their build, on the projects that actually matter to them.
The pieces, and how far each one reaches
| Source | What it gives you | Reach today |
|---|---|---|
| GitHub releases API | New releases, their version tags and release notes, for the repositories you depend on, so a major bump or a breaking-change note can be matched to your rule. | Keyless but rate-capped: unauthenticated requests are limited to about 60 per hour per IP, which sets a sane refresh cadence. A free token raises the ceiling; the cap is stated, not hidden. |
| Status pages and changelogs | Vendor status posts and changelog entries for the hosted APIs that do not live on a public repository, for deprecations and breaking announcements. | Mostly public via RSS or Atom; format varies by vendor, so this is graded verify-at-paste rather than promised. |
The rate caps are real and belong on the page: the unauthenticated releases endpoint allows roughly sixty requests an hour per address, which is plenty for a short watch-list on a quiet cadence and worth a free token for a longer one. Changelog and status-page formats vary, so those are verified when you paste them, never assumed. Feeds across this space are tightening toward keys over time; the page grades honestly and re-checks at paste.
The rule, in their words
- “Only a major version bump, never a patch or a minor.”
- “Anything tagged security, at any version.”
- “Only the five repositories I actually depend on.”
The compound filter is what no shared feed applies: severity AND scope, together. Watching every release across every dependency is what you already ignore. The requirement that a release be major-or-security AND on one of your five repositories is the difference between an alert you act on and a feed you mute.
Unprompted
Your AI
Before you ship, one release crosses your rule: major or security, on your five repositories. The upstream SDK you depend on cut a 4.0.0 two hours ago; the release notes lead with a breaking change to the client initialiser, and 4.0.0 is a major bump, which is your first rule.
That is from the releases feed, read on a quiet cadence within the keyless rate cap, stamped as of two hours ago. It is not in your repository, because it did not happen in your repository.
The release arrived against the developer’s own severity-and-scope rule, from outside the working tree entirely. What to do before shipping is theirs.
Why this beats the dashboard
A dashboard of every release across every dependency is what you already tuned out, and a single-repository watch misses the one you forgot to add. Your rule is severity AND scope, and it surfaces in the turn where you are about to ship. The reading is measured against your own definition of worth interrupting me before it ever reaches you, and it comes from beyond the filesystem, where the search of your machine structurally cannot look.
Getting notified when an API you use ships a breaking change is the public release feeds filtered by your own rule; through Unl a release that meets it arrives beside “major or security, on my five repositories”, stamped with when it was cut and honest about the rate cap, from beyond your working tree, and the rule that judges is yours.
Reads through Unl arrive with measured context — in the presence of the decisions you’ve already settled. The reach lane is live: one box, paste anything. If it speaks MCP, Unl can reach it. Readings arrive unprompted, the data beside the criterion; Unl is a courier, not a warehouse, and keeps only your keys and the frame.
Read further
Questions people ask
What are the rate limits on the release feeds?
The releases endpoint is keyless but rate-capped at roughly sixty requests an hour per IP address unauthenticated, which is ample for a short watch-list read on a quiet cadence; a free token raises that ceiling if you watch more. Status pages and changelogs vary by vendor and are verified when you paste them. The caps are stated on the page because pretending a keyless feed is unlimited would be dishonest.
Why not just watch every release?
Because that is what you already ignore. Your rule is narrow on purpose, a major version or a security fix, on the five repositories you actually depend on, and it is the compound of severity and scope that no shared feed applies for you. The point is to be interrupted only by a release that meets your own definition of worth interrupting.
How does this relate to agentic search of my codebase?
It is the neighbouring case. A coding agent searching your filesystem is the right model for everything in the repository, exhaustive there and silent beyond it. Last night’s upstream release is not in your working tree at all, so no search of the machine can fetch it; it has to arrive from outside, measured against your rule, which is exactly what this construction does.
What this is
Think inside your AI world — you stay in command
Unlimitless (Unl to friends) holds what you've settled, reads what your tools are showing, and catches what's changed out in the world — and hands your AI whatever bears on the work, the moment it's needed, without you asking. The right thing, in front of the model, unprompted, with you in command of the call. So you keep moving toward what you set out to build, on top of everything you've already decided.
It plugs into Claude, Claude Code, ChatGPT and Cursor as an MCP connector. Quick to connect, in a couple of steps.
Unlimitless is open now to invited Alpha. Apply for the Beta waitlist to come in ahead of the full launch:
Alpha is invite-only · free at launch.