Search grounding is now in ThinkReview Cloud 2.2.0
Search grounding is now in ThinkReview Cloud 2.2.0
Reviews can now check the live web — model catalogs, changelogs, and docs — instead of guessing from a training cutoff.
Frontier models ship on a weekly cadence now. Libraries, CVEs, and API IDs move just as fast. A reviewer that cannot look anything up will still sound sure — and still be weeks behind.
ThinkReview Cloud 2.2.0 now enables search grounding on cloud reviews. When a finding depends on what exists today — a model name, a package version, a deprecation — the review can check current sources instead of inventing a story from memory.
This sits on top of the full-context first pass already in 2.1: the repo for what your code does, the web for what the outside world currently is.
The failure mode: confident, and a month late
Here is a real finding from before search grounding. The review flagged a fallback chain as using “nonexistent” model IDs:
The names it called invalid had shipped in the weeks before the review. It even predicted HTTP 404s from the vendor API.
Those models are real. They are selectable in ThinkReview today. The review did not invent random strings. It applied a stale map of “valid IDs,” then wrote a plausible incident: fallback fails, direct calls 404, production outage. That is what a cutoff looks like in a fast market — not nonsense, just last quarter's catalog stated as fact.
With search grounding on, the same review can look up the current vendor docs and walk the finding back:
The correction cites the live catalog: the IDs in the code are stable. It also notes what actually changed — which generation is current, which is previous, and which older ID has been shut down. That last bit is the point. Search grounding is not only “don’t flag real IDs.” It is “check today’s docs before you publish a 404 story.”
What search grounding actually does
Search grounding is not a second product. It is a check the review can run while it reads the diff.
When a claim depends on a fact that lives outside the repository, Cloud reviews can look it up on the live web and ground the finding in those results.
Typical lookups in a review:
- Model and API IDs — does this name exist in the current catalog, or is it a typo?
- Library and framework versions — did this method move, get deprecated, or land in the version you actually pin?
- Advisories and breaking changes — is this CVE, changelog note, or migration still accurate?
The model still has the diff and repository context. Search is for the facts that live outside the repo.
Why code review needs the live web
Stale knowledge is not only a model-name problem. It shows up anywhere the world moved after the checkpoint.
- Model catalogs — vendor IDs and “lite” variants churn. A review that cannot search will flag shipping names as bugs, or miss a real typo because it never knew the new ID.
- SDK surfaces — cloud and payment APIs rename methods between minor versions. Docs beat a frozen memory.
- Security — a “this is fine” take on a dependency can be wrong the week a CVE drops.
- Your own stack's neighbors — agent tools, MCP servers, and cloud function runtimes publish breaking notes faster than any reviewer cutoff.
Full context stops the model inventing bugs from a hunk. Search grounding stops it inventing the state of the industry. You still decide what to keep; the finding is just less likely to be last month's Wikipedia.
Same workflow, grounded reviews
Nothing changes in how you run a review. Open a PR or MR on GitHub, GitLab, Azure DevOps, or Bitbucket, start ThinkReview, and pick a cloud model. Search grounding is on for Cloud 2.2.0 reviews.
Local reviews stay local — they do not call out to the web.
- Use the latest ThinkReview extension, MCP, or API client
- Select a cloud model in Model selection
- Run the review as usual — grounding applies when the model needs a live check
Read the Cloud changelog: ThinkReview Cloud v2.2.0 release notes.
Try it on a moving target
The next time a PR mentions a model ID, a brand-new SDK method, or a CVE from this week, run ThinkReview Cloud 2.2.0. You should see fewer “this API does not exist” comments about things that shipped last Tuesday — and sharper catches when the name really is wrong.
- Chrome / Edge / Brave: Chrome Web Store
- Firefox: Firefox Add-ons
- Microsoft Edge Add-ons: ThinkReview for Edge
- Portal: portal.thinkreview.dev
Cloud 2.2.0: search grounding when the facts live on the web. Install ThinkReview and run it on your next PR.