SoftTech AI Registry
Freshworks · Freshdesk Omni

Freddy AI Agent

Customer-facing AI agent in Freshdesk Omni, configured and deployed through AI Agent Studio, that answers from approved knowledge and executes administrator-configured workflows, including chained authenticated API actions against external systems.

Recorded characteristics

Function
Freshworks documents a structured execution sequence: the agent receives a message from a supported channel; interprets intent using approved knowledge sources and approved business policies; executes configured workflows with conditional logic, Actions Library integrations and confidence-based escalation, working alongside Freshdesk routing, priority and queue management; generates a response governed by configured behavioural instructions covering tone, formatting, compliance language and response boundaries; and either responds automatically or escalates to a human agent while preserving context. Two agent types share this architecture: a chat AI agent for real-time conversations on the web widget, WhatsApp, Instagram, Messenger, LINE and other Freshchat channels, and an email AI agent that answers inbound support email in Freshdesk mailboxes, portals and web forms. Workflows are built from six modular blocks: Trigger, Collect info, Condition path, API action, Custom response and Agent handoff. Workflows can be started by a customer message or by AI Agent events (conversation start, positive or negative feedback, fallback, transfer to a human agent, and out-of-business-hours transfer). An API action sends a request to a configured REST endpoint or MCP server, receives a response and exposes selected fields as workflow variables; GET, POST, PUT, PATCH and DELETE are supported, request bodies and URL path parameters can reference workflow variables, and the output of one API action can be used as the input to a subsequent API action in the same workflow. Freshworks' documented worked example chains Fetch Order Details into a condition on order status and then Initiate Refund using customer, total and payment-method values taken from the prior response. Deployed agents can also interpret customer-sent images (.png, .jpg, .jpeg) and respond with troubleshooting steps or policy links.
Data access
Knowledge grounding is administrator-configured: solution articles, uploaded files (.txt, .docx, .pdf, maximum 35MB, not password-protected), publicly accessible URLs and custom Q&As. Freshworks states the agent cannot learn from private URLs and learns only from static text, not video, animation or screenshots. Documented limits include a maximum of 200 files per agent and per account, a maximum of 10 URLs per agent and 25 per account, and learning up to 3,000 pages from a website; the AI Agent Studio limits table separately lists 50 parent URLs and 3,000 child links per parent, so exact current URL limits differ between Freshworks pages. Operational data access is separate and occurs only through workflows: the Collect info block gathers structured customer input, Condition path rules can reference contact and ticket properties, and API actions retrieve data from the specific endpoints an administrator has configured. Freshworks does not document arbitrary access to account or external data outside configured knowledge sources and configured API actions.
Actions
Can take actions
External actions
Conditional
Human confirmation
Conditional
Permission basis
Separate permissions
Administrative control
Documented administrator controls include: creating, testing, previewing, deploying, archiving and restoring AI agents (archiving a published agent unpublishes and unmaps it); selecting chat or email agent type; name, avatar and primary language; knowledge sources and their removal, restoration and performance review; business context and custom instructions; multilingual support; conversation behaviour (introductory message, tone, response length, fallback handling, feedback, user-detail collection); handover settings covering conversation closure, transfer to a human agent, auto-resolution timing and out-of-business-hours transfer; workflow creation by visual builder or natural-language authoring, workflow drafts, versioning, cloning, deletion and library management; mapping workflows to AI Agent events, with validations that block deleting or unlinking workflows still mapped to events; the API Actions library, including creating, editing, cloning and deleting actions, defining input and output variables, and testing an action against the live endpoint; reusable Authentication configurations; channel mapping and deployment, and for email agents deployment to specific mailboxes, portals and web forms; single-query preview and bulk evaluation testing; and session consumption, performance dashboards, conversation logs and recommended fixes.
Default state
Disabled
Availability
Freshworks documents Freddy AI Agent Studio as Freddy AI functionality in Freshdesk Omni. The API actions documentation states the account must be on Freshdesk Omni or have AI Agent Studio enabled, that administrator access is required, and that at least one AI agent must exist. Chat AI agent setup lists a Freshdesk Omni plan with AI Agent Studio enabled as a prerequisite. Documented system limits include 50 AI agents in total across chat and email per account, 150 workflows per account, 20 workflows per AI agent and 100 nodes per workflow. Natural Language Workflows are documented as available through an Early Access Program requiring Customer Success Manager approval; the rest of the capability is documented as generally available functionality.
Licensing
Freshworks documents AI Agent usage as session-based: a chat session covers all interactions between a customer and the agent within a 24-hour window, and an email session is a 72-hour window from the customer's first email. Session packs are sold in units of 100 sessions for $49 through the billing page, and session consumption is shown in AI Agent Studio. Freshworks also documents Freddy AI as an add-on in the Freshdesk Omni getting-started guide.
External model or provider
Not established. The current Freshdesk AI Agent Studio documentation reviewed does not identify an underlying model provider, model family, model version or any model-selection or routing control available to administrators.
Limitations and uncertainty
Authority exists only where an administrator has configured it: the agent can call only the specific endpoints defined as API actions, using credentials stored in a reusable Authentication configuration (API key, bearer token, basic auth or OAuth 2.0), so this is configured integration authority rather than arbitrary external access or the customer's own permissions. Freshworks lists an integration catalogue spanning payments, e-commerce, shipping, CRM, support and ITSM, project management, HR, identity, messaging, marketing, scheduling, finance and travel, but that catalogue describes what workflows can be built to do rather than proving each operation exists by default. The documented refund example demonstrates configured refund authority in one workflow; it does not establish universal payment or refund authority. Confirmation before sensitive actions is an available workflow design pattern, not a platform-wide guarantee, and Freshworks states that confirmation blocks are skipped for email agents because there is no live conversation, advising safeguards at the API layer instead. Failure behaviour is documented only for chat: an API timeout or failure routes the workflow to the fallback path; no retry, rollback or transaction guarantee is documented. Freshworks states the Trigger block cannot act on system events or detect sentiment. Confidence-based escalation is documented in the architecture description, but a configurable confidence threshold is not established in the pages reviewed. Session history, conversation logs and performance analytics are documented for administrators, but they are not described as immutable audit logs. Freshworks' statements about customer data and AI model training were not established from the primary support documentation reviewed. This record covers the Freshdesk Omni customer-support agent only; Freshservice AI agents, Freddy AI Copilot, Freddy AI Insights and Freshdesk chatbots are separate and are not represented here. MCP is listed as a supported connection option for API actions, but the exact scope of MCP support for this capability is not fully established.

Evidence