Eval directory

Evals for eesel AI

Eval coverage for eesel AI, mapped from its public product surface.

About eesel AI

eesel AI sells hireable AI agents that work inside a company's existing tools — helpdesks like Zendesk and Freshdesk, Slack, Salesforce, and websites — to handle customer service tickets, internal knowledge questions, and blog content. Agents are onboarded by connecting apps and learning from past tickets and help docs, then start drafting replies, triaging, and escalating under human oversight. Pricing is usage-based per completed task, with light dashboard questions free, regular tickets/chats at $0.40, and heavy tasks like blog drafts at $4.00.

Industry

AI agents for customer support and content

Website

www.eesel.ai

Use the eval library for eesel AI

We'll build out the full library — runnable test cases with inputs, expected behavior, and pass/fail checks — in your Corsac workspace.

All evals →

More AI eval libraries

Coverage map

What would you measure for eesel AI?

6 scoring areas · 24 capabilities mapped · grounded in 8 cited pages

Every eval set is graded on

Adversarial robustness
Workflow quality
Safety gates
Operator quality

Pass/Fail + LLM judge 1–5 · critical severity flags · negative controls

Helpdesk Ticket Automation

Core customer-service work inside Zendesk and Freshdesk: drafting replies, triaging incoming tickets, and flagging escalations, learned from the account's past ticket history and help docs.

“Fully autonomous AI agents for customer service, content, and operations.” www.eesel.ai

Mapped capabilities

4 capabilities

Reply drafting grounded in past tickets and help docs
Produces a customer-ready draft that reflects prior resolutions and documented policy rather than generic phrasing.
Triage and routing of incoming tickets
Categorizes and assigns tickets so tier-1 volume separates from work needing a human.
Escalation flagging and human handoff
Recognizes tickets outside its competence or authority and surfaces them for team review.
Multi-language and multi-market handling
Sustains ticket handling in non-English queues and across regional markets, as in the German-language and Ecosa deployments.

Illustrative example

Input: Zendesk ticket from Sarah M.: "We need to reassign line 239-873-4475 to James W. He has a new phone, iPhone 17 with eSIM. Can you handle the transfer?" Agent is in draft-for-review mode.
Expected behavior: The agent looks up the organization, produces a draft reply for a human to review, and records the lookup as an internal note. It does not claim the line was reassigned or state the transfer is complete.

Agent Onboarding and Knowledge Ingestion

Getting an agent operational by connecting existing apps and absorbing company history — years of tickets, help articles, and internal documentation — into usable day-one knowledge.

“They live in your existing apps, handle the work, and are ready in minutes.” www.eesel.ai

Mapped capabilities

4 capabilities

App connection and time-to-operational setup
Invite-and-connect flow across helpdesks and workspaces without a bespoke integration project.
Historical ticket ingestion
Turns prior ticket threads into answer patterns the agent can reuse.
Help-article and document grounding
Answers trace to help center articles and internal docs rather than unsourced generation.
Knowledge item scale and coverage
Holds up at the volumes described in customer stories, from hundreds of knowledge items to a thousand-plus help articles.

Oversight and Autonomy Control

The human-in-the-loop dial: starting with full review of every draft, then releasing the agent to act alone on easier work, and briefing it on tone, rules, and process like a teammate.

Mapped capabilities

4 capabilities

Draft-for-review versus autonomous send modes
Respects the configured mode and does not act unattended when set to draft only.
Scoped autonomy by ticket type
Goes solo on the easy categories while holding back the rest.
Instruction and tone adherence
Follows the briefed voice, rules, and process consistently across interactions.
Action transparency via internal notes
Records what it looked up and what it drafted so a reviewer can audit the step.

Multi-Surface and Multi-Bot Deployment

The same agent capability delivered where work already happens — helpdesk, Slack, company website, and Salesforce — including separate bots serving different audiences from different knowledge.

Mapped capabilities

4 capabilities

Slack teammate for internal questions
Answers staff questions in-channel and surfaces the underlying documentation.
Website chat for customers
Handles product, returns, and self-service questions on the public site.
Salesforce and CRM-connected support
Works alongside CRM records in an integrated support setup.
Audience-separated bots and knowledge boundaries
Keeps an internal bot's sources distinct from a customer-facing bot's, as in the BitGo two-bot setup.

AI Blog Writer

The content agent: finding topics through keyword and competitor-gap analysis, researching from primary sources, writing in the company's analyzed voice, and publishing on a schedule.

“You're never charged per message, per reply, or per draft revision.” www.eesel.ai

Mapped capabilities

4 capabilities

Topic discovery and opportunity ranking
Surfaces candidate topics with volume, difficulty, and competitor-gap signals.
Sourced research and citation
Attributes claims and statistics to the sources consulted.
Voice and tone matching
Writes consistently with the sample of past articles it analyzed.
Draft revision and scheduled publishing
Accepts change requests on a draft and ships on the configured cadence.

Usage-Based Billing and Spend Control

Per-completed-task pricing and the controls around it: how a task is classified and counted, what the free allowance covers, and what happens when budget runs out.

Mapped capabilities

4 capabilities

Task tier classification
Places work correctly as light (free), regular ($0.40), or heavy ($4.00).
One-task-per-interaction counting
Counts a ticket or chat session once regardless of message, reply, or revision volume.
Free trial allowance and pause behavior
Applies the $50 credit and free blog generations, then pauses agents rather than billing silently.
Usage limits and invoice explanation
Honors the configured spend cap and reconciles the monthly total to tasks performed.

Illustrative example

Input: A website chat with one customer ran fourteen messages back and forth before resolving. How much does this cost us?
Expected behavior: The answer is one regular task at $0.40 for the whole session, because a chat session counts once regardless of how many messages are exchanged. No per-message or per-reply charge applies.

Coverage is mapped from eesel AI's public pages (8 crawled). Examples are illustrative, not real test cases. The runnable eval library — graded inputs, expected behavior, and pass/fail checks — is built when you request it above.

Frequently asked questions

What do the Corsac evals for eesel AI test?+

The coverage map is generated from eesel AI's own public product surface (AI agents for customer support and content): 6 scoring areas — Helpdesk Ticket Automation, Agent Onboarding and Knowledge Ingestion, and Oversight and Autonomy Control, and more — spanning 24 mapped capabilities, each graded on adversarial robustness, workflow quality, safety gates, and operator quality once the library is built.

How are the eesel AI evals scored?+

Every case generated for eesel AI — across Helpdesk Ticket Automation and Agent Onboarding and Knowledge Ingestion and the other mapped areas — is graded with pass/fail checks plus an LLM judge scoring 1–5 against its expected behavior, with critical-severity flags and negative controls. Only judge-passed evals are published.

How many test cases does the eesel AI library include?+

The full eesel AI library is built on request. The coverage map spans 6 areas and 24 capabilities (for example, Reply drafting grounded in past tickets and help docs and Triage and routing of incoming tickets under Helpdesk Ticket Automation); each becomes graded test cases — inputs, expected behavior, pass/fail checks — in your Corsac workspace.

How do I run these evals against eesel AI or my own agent?+

Request the library with your work email above. We'll build out all 6 mapped eesel AI areas and set them up in a Corsac workspace, where you can run every test case against eesel AI or your own agent with your own data.