Zoom Acquires Onyx: The Open-Source Search Layer It Didn't Have
SoWhatNowWhat — Signal
• Impact Threat Level: Medium — a genuine new capability for Zoom, hard to copy quickly given three years of open-source community building, but it sits next to core UC/CX/CCaaS territory rather than inside it, and the deal terms are undisclosed. Decision: M&A • THE "WHAT": Zoom bought Onyx, an open-source, self-hosted search tool. It has over 30,000 GitHub stars, handles a million queries each week, and serves clients like NASA, Nebius, Ramp, and UC San Diego. • THE "SO WHAT?": Zoom gains an in-house search system it did not build from scratch, which already holds the trust of safety-focused buyers. The surprising detail is what Zoom promises to leave alone. Onyx will remain open source, flexible with other models, and available to run locally, which is rare when large firms buy smaller software companies.
Summary
Zoom bought Onyx this week. We found out through an email from Zoom's Analyst Relations team, and a post on LinkedIn from Onyx co-founder Chris Weaver. This happened during the same week XD Huang, Zoom's CTO, spoke at HumanX Amsterdam. Onyx was set up in 2023 to help companies run enterprise search on their own servers, so they do not have to hand sensitive data to outside parties. The business is now part of Zoom AI Services.
Weaver shared his own figures, including an 18-person team, over 30,000 GitHub stars, more than a million queries each week, and clients such as NASA, Nebius, Ramp, and UC San Diego. Zoom says four main ideas guide its AI work: Open, Sovereign, Multilingual, and Trusted. Sovereign is the key factor here. Zoom claims Onyx helps large firms with strict data laws and local server requirements, which basic cloud tools simply cannot satisfy.
Zoom is staying quiet about the financial details, the timeline, and how Onyx will actually fit into its broader AI platform. Meanwhile, Weaver maintains that the software will remain open source, flexible with different models, and available for local hosting. That is a clear promise, and the large community on GitHub will certainly spot any changes. Right now, though, we only have their word for it.
What Happened: The Zoom Onyx Acquisition
The acquisition: Zoom has bought Onyx. Its own analyst relations team confirmed it, and so did Onyx co-founder Chris Weaver, separately.
What Onyx brings: An open-source, self-hosted search and knowledge platform, running since 2023, with real production use: over a million queries a week, for customers including NASA, Nebius, Ramp, and UC San Diego.
The stated rationale: Zoom says the deal is about data sovereignty and on-premises deployment, needs its existing AI Services stack couldn't meet alone.
What Zoom is promising: Onyx stays open source, self-hostable, and model agnostic after the deal. That's a public promise. It hasn't been tested yet.
Why the Self-Hosted Commitment Is the Real Story
Zoom has done this before. It bought BrightHire and Common Room in the past year too, each one moving the business into new opportunities. Onyx pushes Zoom into enterprise search and data control, which fits what the company says about the purchase. It wants to reach big firms with strict local server rules, which cloud systems just can't satisfy.
The key detail to watch is the self-hosting promise. Most tech deals pull a product into the buyer's own cloud and pricing setup within a year or two. Zoom saying it will not do that, in public, on day one, is either a real promise to the open-source community, or a line that changes once the plans merge. Either way, it is on the record, and the 30,000 people on GitHub will see if it changes.
Where This Actually Shows Up
Inside Zoom, the fit makes total sense. Onyx already sits on Zoom's AI Services page as one of three main API tools, alongside Speech and Language, so this is not a random add-on, but central to the strategy. ZoomMate, Virtual Agent, Contact Centre, and the AI tools like Canvas, Sheets, Slides, and Paper all need the same thing, which is a clear answer grounded on real company data.
It is worth being clear here, as Zoom already has plenty of tools. Meetings, Chat, Phone, Mail, Calendar, Whiteboard, Canvas, Sheets, Slides, and Paper cover most of what Microsoft 365 and Google Workspace do, and Zoom bought its way further into that market this year with BrightHire for hiring, and Common Room for sales leads. Onyx follows the same idea. It is another bought-in feature added to a product set Zoom is actively building.
Onyx matters most for the many buyers using Zoom for calls and contact centres, but relying on Microsoft 365 or Google Workspace for emails, files, and documents. Onyx connects to Slack, Google Drive, GitHub, Confluence, Salesforce, and Notion. For these users, it lets Zoom's AI search through files outside its own system, not just the meeting notes it holds. That gives Zoom a real edge over search tools locked to a single provider, and it matches the pitch Glean already makes to the same buyers.
The caveats still stand. A company already paying for Copilot or Gemini already has an AI layer over the same Microsoft 365 or Workspace data, so adding a second search tool over the exact same data is a tough sell, unless Zoom is clearly cheaper or better, and nothing shared so far proves that. Microsoft and Google also have every reason to restrict API access if a rival uses it to build a competing AI service on their platforms. Plus, managing two separate secure indexes over sensitive files doubles the work for security teams, rather than cutting it in half.
What Buyers Should Still Verify
Onyx guides show three ways to set up a link: open to everyone, limited to a list chosen by an admin, or copied from the original system. That third option, which handles the real security settings, currently works for Google Drive, Confluence, Jira, and Slack. Anything else added to the index does not offer that same clear promise. Independent tests also show that full security syncing needs the paid Enterprise version, and is missing from the free option.
This matters a lot, especially for the exact buyers Zoom wants to target. These are large firms with strict data rules, the ones dealing with board notes, financial statements, or details kept private during quiet periods.
Zoom's own AI tools actually have more privacy controls than some might expect. Zoom says it does not use customer content to train models, offers an option to delete transcript data instantly, and lets managers automatically remove sensitive items like names, bank details, passport numbers, and email addresses from summaries. Admins can also set custom rules to ban specific phrases. For example, Zoom can block the phrase "internal use only" from ever showing up in a summary. That sort of setting can stop private financial news from leaking into a brief in the first place.
Two big problems may remain beneath the surface. First, both the redaction feature, and the custom rule, need a support ticket to turn on. Neither option works automatically, so a firm that has not asked for help has no cover, no matter how good the feature looks in sales decks. Second, the redaction list focuses on personal information, not confidential business data, so it will miss unreleased financial details, and secret project names, unless someone adds a rule by hand.
None of this proves a clear fault in what Zoom or Onyx builds. It simply raises clear questions from their own published documentation that are worth checking. These include which tools keep original safety rules inside Zoom, whether blocking settings work straight away or need a support request, and if ZoomMate searches only the files a user has rights to view, or something wider. It is better to ask Zoom directly, rather than guessing.
Get the deeper read.
Source: Chris Weaver, Onyx co-founder, via LinkedIn, 22–24 Sept 2026 — https://lnkd.in/p/eU_zWWmz
Documentation referenced:
About the analyst: Tim Banting, 20 years in UC/CX market intelligence (Microsoft, Cisco, Omdia, GlobalData).
