Catalyst 3.0 is Zoho’s bet that writing software is no longer the bottleneck. For SMBs and agencies already living in Zoho, the release is a way to let coding agents provision, wire, and deploy without assembling a second cloud.
Zoho launched Catalyst 3.0 in early September 2026 as an “agent-ready” full-stack cloud. The company has been repeating one point in briefings and regional rollouts ever since: models can write a lot of the code now, and getting that code into a running, billed, authenticated application is still the slow part. Catalyst is trying to sit in that gap. It keeps the serverless platform Zoho has been building since 2021, then adds the interfaces an AI coding agent needs to use that platform without a human clicking through a console for every table, auth setting, and deploy.
Anand Nergunam, Zoho’s global vice president for revenue growth, put the problem in plain language. Writing code has become simple and quick. Taking that code live can still take months. Catalyst 3.0 is aimed at in-house tech teams, system integrators, custom software shops, independent developers, founders, and students. The commercial pitch is usage-based infrastructure with no per-user seat tax and no egress fee, plus a monthly free tier and introductory credits. Students get the full product for non-commercial use.
What “opening up for agents” actually means
Three pieces matter, and they work as a chain rather than as separate features. Agent Skills are machine-readable files that tell a coding assistant which Catalyst services exist, how they should be wired, and what “good” looks like on this platform. Zoho has published those skills for the tools people already have open: Claude, Cursor, Gemini, GitHub Copilot, Codex, and Devin Desktop. In Zoho’s own tests, the same model on the same task jumped from roughly 35 percent end-to-end completion without skills to more than 90 percent with skills and MCP. Human interruptions and retries dropped with them. Those numbers are vendor tests, so treat them as a demonstration of the mechanism, not as your future SLA.
The second piece is a live MCP server. Model Context Protocol is the common way agents call tools. Catalyst’s server lets Claude, Cursor, Codex, Gemini, and similar clients create tables, add columns, wire authentication, and deploy against real infrastructure from the editor. The agent is not only generating a README that tells you to log into a dashboard. It can perform the platform action, then keep going. The third piece is a non-interactive CLI. Agents can run a multi-step Catalyst workflow without stopping to ask a human “are you sure?” at every prompt. That is how you get from a chat transcript to a deployed preview without fifteen context switches.
Zoho also open-sourced the skill files, SDKs, CLI, and the Slyte JavaScript framework so developers can inspect and extend them. The platform underneath stays proprietary. That split is useful. You can read what the agent is being taught. You cannot fork Catalyst itself. For an agency that has to explain tooling choices to a client’s IT team, being able to show the skill files is more reassuring than a black-box “AI cloud” slide.
The platform around the agents
Catalyst is not only a connector layer. It is a full-stack cloud with more than 30 services on one bill: functions, hosting, relational and NoSQL data, object storage, auth, cache, queues and signals, frontend hosting, browser automation, and Zoho’s own ML and Zia pieces. The 3.0 story adds zero egress as a structural guarantee rather than a promo rate. For a small product that serves files, images, or API responses to users all day, egress is often the line item that makes a cheap compute bill feel expensive. Removing it changes the spreadsheet more than another five percent off functions.
Governance is the other half of the 3.0 pitch, and it is the part SMBs skip until something goes wrong. Dev and production stay decoupled. Code is not promoted to production just because an agent finished a task. The agent can log in with its own scoped permission set. Destructive commands are disabled in non-interactive mode. Application logs, platform logs, and MCP tool-call logs are kept so you can reconstruct what the agent did and when. Changes after launch are versioned, attributable, and reversible. None of that is glamorous. All of it is what you want when a contractor, a junior, or an agent is allowed to touch infrastructure.
Nergunam’s comment on control is the right frame for buyers who have been burned by “just let the AI do it.” Good developers want to know which components are being used, how they are configured, and whether the architecture is sound. Catalyst’s intended loop is: the agent proposes, a human approves the architecture, MCP and the CLI execute, production still waits for a promotion. If your team cannot staff that review step, the platform will not save you. It will only help you ship the wrong app faster.
Where this is actually useful for SMBs and agencies
The highest-leverage case is a company that already runs on Zoho and needs custom software around it. CRM, Books, Desk, and the rest of the suite cover a lot. They do not cover every client portal, approval flow, field app, or partner integration. Those extras usually land on a freelancer or a small SI, who then has to pick a cloud, wire auth, stand up a database, and remember to send an invoice for the babysitting. Catalyst gives that person a place to land the custom piece on the same vendor family, with an agent that already knows the services.
Agencies should look at it as a delivery lane, not as a religion. If a client wants a prototype this week and a maintained app next quarter, the expensive part is rarely the first 400 lines of frontend. It is auth, environments, hosting, logs, and the 2 a.m. “why is this preview broken” thread. An agent that can create the datastore, attach login, and deploy a preview from Cursor or Copilot compresses that middle. You still own QA, security review, and the conversation about what happens when the founder’s cousin becomes an admin.
System integrators get a staffing argument that is more honest than “AI replaces developers.” If infrastructure and DevOps after coding shrink, the same two engineers can run more projects. That only works if the projects fit the platform. A greenfield internal tool, a customer portal, a lightweight mobile backend, or a Zoho extension is a fit. A specialized data platform with exotic networking is not. Use Catalyst where the service catalog already has the parts. Do not invent a reason to rebuild a mature AWS estate just because skills files exist.
Founders and very small product teams get a path from prompt to preview that does not require a platform engineer. Zoho has cited a Catalyst-built app that reached about two million users after a three-week build by one developer. That is an anecdote, not a template. The usable point is the shape of the bet: if the agent knows the cloud, a competent generalist can ship farther than when every service was a new vendor and a new dashboard.
Cost, data residency, and the boring constraints that decide deals
Pricing is usage-based. There is no per-user fee and no charge for moving data out. Zoho has been offering a monthly free tier that renews, budget alerts, spending limits, and launch credits on the order of $250 or Rs 15,000 depending on the market. Students get the full feature set for non-commercial use. For an SMB, the commercial question is not “is there a free tier.” It is whether you can put a monthly cap on an agent that is allowed to provision resources. If you cannot, pause the MCP connection until you can.
Data residency decides some deals and is noise in others. Catalyst runs on Zoho infrastructure that already serves more than 150 million users, with data centers in India, the United States, and Europe, plus local residency called out in Middle East launches for Saudi Arabia and the UAE. If clients will not put customer data in a US-only region, that map matters more than another agent demo.
How to start this week without boiling the ocean
Install the Catalyst skills in the coding agent you already use. Zoho documents a one-line add for several clients, including an npx install for Gemini and marketplace listings for Claude Code, Codex, and Cursor. Connect MCP against a sandbox project, not production. Ask the agent to propose an architecture for one real internal tool: a lead intake form that writes to a datastore, sends a notification, and sits behind Zoho login. Approve or reject the architecture in writing before you let the CLI deploy anything.
Measure the trial the same way you would measure a junior contractor. Did the agent pick the right services. Did it create secret sprawl. Did you understand the logs. How many times did a human have to stop it. If that loop is clean, point it at a client preview next, still off production. If the loop is sloppy, you learned that the bottleneck in your shop is review capacity, not cloud consoles. That is still a useful result.
Catalyst 3.0 will not make software cheap by itself. It removes a class of glue work that small teams were never staffed to do well, and it teaches the agents those teams already pay for how to use one coherent platform. Used with a human still on the promotion button, that is a practical upgrade. Used as an excuse to skip design, security, and ownership, it is just a faster way to create another app nobody wants to maintain.