Freddy AI agents: features, setup, pricing, and a CLI alternative

What are Freddy AI agents?

The useful buying question is not just whether Freddy can answer a question. It is which support job you want it to own, what information it can access, and what happens when it cannot finish.

I would separate three offerings before comparing features:

This distinction matters when a demo moves from answering a help-center question to changing an order. A useful answer requires the right knowledge. A successful action also requires the right integration, inputs, permissions, and failure handling. A confident response is not evidence that an order was changed.

What Freddy AI agents can do

Build customer-support behavior in AI Agent Studio

Freshworks' AI Agent Studio guide describes an admin workflow: create the agent, add knowledge, configure workflows and instructions, test, then map it to a channel. Workflows can be built directly or started from library templates. Check the specific action you need rather than assuming every connected app supports every operation.

The guide lists public URLs, files, solution articles, and custom Q&As as knowledge inputs. That gives teams several ways to supply answers; it does not mean a private document URL will be readable. For source limits and scope, see the knowledge sources guide.

The knowledge panel shows learning status for connected URLs. Source: Freshworks' knowledge configuration guide.

Freshworks also provides testing and a shareable preview. There is an important limit: preview does not perform agent transfers for unanswered questions or requests for a human. Test that handoff in the deployed channel rather than marking it passed after a preview conversation.

Keep Freshservice's employee-support setup separate

The Freshservice overview documents Freddy AI Agent for Enterprise, across Slack, Microsoft Teams, email, and the support portal. Slack and Teams require the corresponding ServiceBot installation before configuration.

It also lists enterprise search across SharePoint, Google Drive, and Confluence. So a blanket claim that Freddy cannot use knowledge outside Freshworks would be wrong. The better question is whether the particular source, permissions, and employee audience you need are supported in your setup.

Freshservice includes 1,200 sessions per year per Enterprise license, with a unique-user, 24-hour session definition. That is a Freshservice entitlement, not the customer-support session pack described below.

Freddy AI agents pricing

As checked on September 8, 2026, Freshdesk Omni pricing lists additional Freddy AI Agent usage at $49 per 100 sessions, after 500 introductory sessions available once per account. Those 500 are not a recurring monthly allowance.

Its FAQ defines chat sessions as a 24-hour window and email sessions as a 72-hour window from the customer's first email. Pack validity follows the payment cycle. Include the underlying helpdesk subscription and any separate add-ons in your budget.

For eesel, the current pricing documentation uses tasks rather than the old monthly interaction bundles:

eesel bills tasks regardless of outcome. A spend cap can pause new work, but it does not make a usage-based plan a fixed bill or prove the replies were correct. Avoid comparing only the two unit prices: the billing units differ, and human review, escalations, and coverage affect the result.

Evaluate a Freshdesk teammate through eesel CLI

eesel's helpdesk teammate is another option for handling support in your existing queue. The CLI makes the setup accessible from the tools a technical teammate already uses. A person can type commands, a script can collect output, and Claude Code, Cursor, or Codex can run a scoped setup check.

The terminal and dashboard operate the same eesel agent and workspace. You are not building a second support bot inside your coding agent. For example, Claude Code can inspect the eesel agent's integrations and instructions, then report what is missing before you enable a Freshdesk pilot. The CLI does not administer Freddy.

1. Connect the knowledge and choose the test agent

The Freshdesk integration guide distinguishes public-help-center Quick start from Full connection. The latter uses your subdomain and Freshdesk API key for tickets, canned responses, triggers, and actions. Put the key in the secure form provided by eesel, never in a chat message or pasted command.

For this example, select or create a separate test agent in the dashboard. Connect only sources approved for the pilot, check their sync status, and replace TEST_AGENT_ID below with that agent's ID. A separate agent helps keep configuration distinct; it does not isolate a live Freshdesk account if you connect one.

With Node.js 18.17 or newer, these documented commands inspect the setup:

npx @eesel/cli login
npx @eesel/cli whoami
npx @eesel/cli agents
npx @eesel/cli --agent TEST_AGENT_ID status
npx @eesel/cli --agent TEST_AGENT_ID integrations
npx @eesel/cli --agent TEST_AGENT_ID instructions
npx @eesel/cli --agent TEST_AGENT_ID automations

Check the workspace before proceeding. Then check what knowledge is actually available, which standing instructions apply, and whether any automation is already enabled. Connecting a source and enabling a ticket workflow are separate steps, but an existing automation may already be running.

2. Test the decision before enabling ticket actions

For a first test, use a fictional support scenario with a clear expected answer. Keep write actions off, or subject to approval, in the agent's actual settings. A prompt saying "do not send" is guidance, not a permission boundary.

Here is an illustrative policy test. It is not a command sequence I ran against a customer account:

npx @eesel/cli --agent TEST_AGENT_ID new --name "duplicate-charge-pilot"
npx @eesel/cli --agent TEST_AGENT_ID chat "Fictional support test. Policy: before investigating a duplicate charge, ask for the order number and the two charge dates. Never request a full card number or promise a refund before verification. Customer: I was charged twice, please refund me. Draft a response here for a human reviewer and explain what information is missing. Do not post to Freshdesk or change anything."

The expected response asks for the missing details without inventing a refund decision. Follow with a customer who supplies only one charge date, then one who asks to send full card details. Those variations test whether the agent follows the policy beyond its easiest case.

The policy in this prompt is test context. It does not replace persistent instructions or your approved knowledge. For the next stage, configure your real policy and test against it, including cases not used while editing the instructions.

The new command starts a conversation, not a new agent. Ordinary chat commands continue the current conversation. Consult the CLI reference for current syntax; chat can invoke tools and run billable work, so keep tests small and permissions explicit.

3. Check what happened, then try a private-note pilot

After the test, inspect the records:

npx @eesel/cli --agent TEST_AGENT_ID activity
npx @eesel/cli --agent TEST_AGENT_ID approvals
npx @eesel/cli billing

Single results are JSON and lists use one object per line. That makes it practical for a coding agent to summarize the result or for a script to retain a review record. Treat held approvals as proposed actions, not completed work. Billing shows current state, not a forecast of savings.

Before allowing a real workflow, use a test ticket in an account you are authorized to modify. For eesel's Freshdesk integration, an internal-note action leaves a draft for your team; a ticket-reply action sends to the customer. Keep customer replies and unrelated write actions off while testing private notes. Leave ticket merging off unless deliberately required, since Freshdesk cannot undo it.

A terminal answer alone cannot establish the last three. You need the actual ticket and activity record. Also, CLI --dry-run previews a server write call; it is not a simulation that proves answer quality or predicts ROI.

Choosing between Freddy and an eesel teammate

Freddy is worth evaluating in the Freshworks product you already use, with the native knowledge and workflow features that match your job. Compare it with an eesel teammate using the same approved policy and difficult support cases. Measure what each actually did, including escalation quality and human correction, rather than comparing demo fluency.

The eesel dashboard is another way to work with the same teammate available through the CLI. This example shows a Zendesk setup, not the Freshdesk pilot above.

For a Freshdesk team with technical colleagues involved in rollout, eesel CLI gives those colleagues concrete work: inspect source readiness, check instructions, run bounded questions, and review activity. Your support team can keep reviewing the same teammate in the dashboard and helpdesk.

Try eesel and start with one support policy and one private-note workflow. Expand only after the answers and actual ticket behavior meet your review criteria.