Freshworks

Authenticated action into external systems through a documented, typed mechanism. Freshworks publishes AI Actions apps, which expose typed callable operations to AI Agent Studio for use inside agent workflows to perform tasks such as updating records and resetting passwords, backed by an Actions Library the agent calls at runtime.

Knowledge and tool connections span Google Drive, SharePoint and Confluence, agents call third-party tools including Notion and Workday, and Freshservice connects to external AI tooling over the Model Context Protocol. Public REST APIs cover product entities including tickets, contacts and agents.

Sourcedevelopers.freshworks.com/skillsread 2026-09-05

Action runs through workflows the agent calls, which is the documented model rather than a claim about autonomy: when a conversation needs to do something, reset a password, approve access or raise a service request, the agent calls a workflow, and workflows execute with conditional logic, Actions Library integrations and confidence-based escalation alongside the helpdesk's routing, priority and queue management. Published end-to-end journeys span IT from access provisioning through password resets to device setup, and HR from onboarding through leave requests to payroll queries, handled across systems rather than within one.

Sourcesupport.freshservice.com/support/solutions/articles/50000012643-introduction-to-ai-agent-studio-in-freshserviceread 2026-09-05

A maintained retrieval structure the customer assembles and points the agent at.

Freshworks documents that each agent is grounded in the customer's own service data plus any additional knowledge sources connected to it, naming files, URLs, the knowledge base, the service catalog and connected apps including Google Drive, SharePoint and Confluence, and that when a user asks a question the agent retrieves relevant content and composes a grounded answer.

Agents, knowledge sources and workflows are linked through explicit configuration, so which corpus serves which agent is a customer decision. Answers are constrained to approved knowledge sources under approved business policies.

Sourcesupport.freshservice.com/support/solutions/articles/50000012643-introduction-to-ai-agent-studio-in-freshserviceread 2026-09-05

Controls sit ahead of the action and are documented rather than asserted.

The agent interprets intent using approved knowledge sources against approved business policies to reduce the risk of fabricated information, generates responses within configured behavioral instructions covering tone, formatting, compliance language and explicit response boundaries, and executes workflows with conditional logic and confidence-based escalation, so a low-confidence path hands to a person rather than proceeding.

AI Agent Studio supports localized role-based access control, granting access through roles rather than restricting it to account administrators, and Freshworks publishes deployment guardrails alongside it. The Freshservice AI Agent Studio product page marks workflow versioning and full auditability as coming soon.

Sourcecrmsupport.freshworks.com/support/solutions/articles/50000011711-introduction-to-ai-agent-studioread 2026-09-05

The security page describes role based access through IAM with segregation of duties, two factor authentication and end to end audit trails, displays ISO/IEC certification and AICPA attestation badges, and links the Freshworks Trust Center for certifications and reports; the Freshdesk support documentation defines seven default roles (Agent, Supervisor, Administrator, Account Administrator, Freddy AI Copilot User, Ticket Collaborator, Analytics Collaborator) and custom roles on Pro and Enterprise.

Sourcefreshworks.com/securityread 2026-09-30

Performance reporting is documented; run reconstruction is not. Freddy AI Insights delivers proactive alerts, root cause analysis and performance intelligence to service leaders, and Freshworks publishes performance analytics and gap detection showing how AI performs across topics, resolutions and escalations with guidance to close knowledge or workflow gaps.

This is service and quality analytics rather than a step-level decision trace, tool-call log or retrieval record for an individual agent run. Freshworks' own pages disagree on whether audit trails are live: the Freshservice AI Agent Studio page marks full auditability and workflow versioning as coming soon, while the May 2026 launch post lists auditability among shipped governance, so confirm before relying on it.

Sourcefreshworks.com/freshservice/ai-agent-studioread 2026-09-05

No memory layer with a stated scope or lifetime is documented. The published architecture is agents reasoning over connected knowledge sources and acting through workflows, with context assembled at question time from Freshservice data and connected apps, which is retrieval rather than state the agent retains between conversations. The only persistence unit Freshworks names is the session, and sessions are a metering construct that does not carry over between billing cycles rather than a customer memory a buyer can scope or delete.

Sourcecrmsupport.freshworks.com/support/solutions/articles/50000011711-introduction-to-ai-agent-studioread 2026-09-05

Freshworks runs regional data centers in the United States, the EEA, the UAE, India and Australia, and the customer chooses their preferred data location when signing up, with productive and backup instances deployed in the same region and a documented migration path through support for regulatory requirements. The subprocessor page states data resides in the corresponding facility based on the controller's location or plan choice, and Freshcaller call data can be pinned to the same regional center.

Sourcefreshworks.com/privacy/data-hostingread 2026-09-05

Both building and adopting are offered. Freddy AI Agent Studio lets teams create agents in a no-code workspace or start from domain-specific templates, and Freshworks ships named role-scoped agents a buyer puts into service rather than authors: IT agents covering access provisioning, password resets and device setup, HR agents covering onboarding, leave requests and payroll queries, and a dedicated Email AI Agent for high-volume email support. Freshworks offers the choice to build, customize or deploy bespoke Freddy AI Agents.

Sourcefreshworks.com/freshservice/ai-agent-studioread 2026-09-05

Work reaches the agent across the surfaces employees and customers already use: agents deploy into Microsoft Teams, Slack and employee or self-service portals, answer in chat and email, and Freshworks ships a dedicated Email AI Agent automating high-volume email support with context-aware multi-turn replies, alongside a native voice channel added in the April 2026 release. Freshworks states agents deploy across integrations, channels, knowledge sources and languages to deliver consistent service across every channel employees use.

Sourcefreshworks.com/theworks/company-news/may-2026-launchread 2026-09-05

No model selection, routing control or bring-your-own-model option is exposed to the customer. Freshworks documents the agent lifecycle, knowledge configuration, workflow execution and session metering without naming a model or offering a choice of one, and positions Freddy AI as trusted domain-specific AI grounded in enterprise context rather than as a model-flexible layer.

Sourcecrmsupport.freshworks.com/support/solutions/articles/50000011711-introduction-to-ai-agent-studioread 2026-09-05

A full public developer platform: the Freshworks Development Kit CLI, App SDK at versions 2.3 and 3.0, serverless compute with scheduled and product events, server method invocation and request proxying, a managed data store, REST APIs with regional endpoints, and a Marketplace distributing apps to a stated 75,000 customers. Integrated MCP servers manage the custom app lifecycle, and an App SDK surface lets apps expose AI Actions that Agent Studio agents invoke.

Sourcedevelopers.freshworks.comread 2026-09-05

AI Actions carry a documented local testing step during configuration and a Test API surface inside Agent Studio, so the customer validates each action with input mapping and output configuration before an agent invokes it in production. This tests the components an agent calls rather than the agent itself, with no conversation simulation, scenario suite or scored evaluation documented, which is a quality gate rather than an evaluation product.

Sourcedevelopers.freshworks.com/docs/app-sdk/v3.0/product/whats-newread 2026-09-05

No browser control, hosted session or computer use appears in Freshworks product or developer documentation. Agents act programmatically instead, by calling workflows, which invoke typed AI Actions and Actions Library integrations against connected systems, with public REST APIs and Model Context Protocol connections carrying the rest.

Sourcedevelopers.freshworks.com/skillsread 2026-09-05