What’s There
Curated agents from Zoho and partners across functional categories: Sales Development (lead enrichment, sequence outreach, meeting booking), HR (onboarding orchestration, leave-request triage, expense report routing), Customer Support (ticket triage, response drafting, escalation prediction), IT Help Desk (password reset, asset assignment, common-issue resolution), Inventory (reorder agent, stock-out alerter), Logistics (shipment tracking, delivery exception handler), Finance (invoice matching, payment chase, expense categorization). The catalog launched with ~40 agents in March 2026 and is growing through partner submissions.
Deployment Model
Install an agent into your Zoho apps from the Marketplace listing. Adjust prompts, data mappings, and escalation rules in the agent’s configuration panel. The agent runs against your data with your policies via Zia’s permission framework — it acts as a service account scoped to the modules you grant it. Turnaround is hours, not weeks, for templates that match your process; 1-2 weeks if customization is significant.
Configuration Surface
Each marketplace agent exposes a structured configuration: data scope (which modules and views the agent can read), action scope (what it can write or trigger), prompt overrides (replace default prompts with your domain language), escalation rules (when to hand off to a human), notification channels (which Cliq channel or email gets summary digests), and credit budget (cap monthly executions). Lock down action scope before going live — a sales development agent with write access to Deals can move records into wrong stages if the prompt drifts.
Customization Limits
Pre-built agents cover common patterns. They expose configuration knobs for the most-tweaked decisions but not every internal logic step. Highly custom workflows still need Agent Studio (build from scratch) or bespoke development on Catalyst. Common signs you’ve outgrown a marketplace agent: you’ve forked the prompt three times, you’re routing 40%+ of cases to human escalation, or you need branching logic the agent’s UI doesn’t expose.
Cost Posture
Agent execution bills against Zoho credits. Pricing varies by agent — simple lookup agents run cheaply (low single-digit credits per execution), complex multi-tool agents cost more (10-50 credits per execution depending on LLM call count and tool calls). Monitor consumption under Zoho Billing → Credit Usage daily for the first month after deploying a new agent. A marketplace agent that does its job cheaply is an unqualified win; one running up the bill on marginal outcomes needs re-examination.
Vetting Process
Zoho-published agents are vetted for security (data handling, permission boundaries) and functional correctness. Partner-published agents go through a lighter review — read the partner’s documentation, check the data scope it requests, and pilot before broad deployment. The Marketplace listing shows install count, average rating, and recent reviews; agents with under 50 installs and no recent reviews carry risk.
Common Failure Modes
Over-broad data scope at install — an agent requests “all CRM modules” by default; restrict to what it actually needs. Prompt drift where the agent starts handling cases outside its trained scope and produces low-quality outputs. Credit blow-out when a chatty agent loops on edge cases — set a hard monthly cap, not a soft one. No human-in-the-loop for first 100 executions — review every output for the first batch to confirm behavior matches your process.
What to do this week
Browse the Marketplace for one agent matching a high-volume process (e.g. ticket triage in Desk), install in a sandbox, run 20 cases through it manually, and review every output against your team’s expected handling before considering production.
