Eval directory

Evals for DRUID AI

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

About DRUID AI

Druid AI is an enterprise platform for building and orchestrating AI agents that handle business-critical customer and employee interactions across chat, SMS, voice, and email. It ships pre-built agents and industry templates for healthcare, higher education, banking, insurance, retail, HR, and IT helpdesk, integrating with existing systems like Jira and RPA tooling. Pricing is custom and quote-based, and the vendor emphasizes LLM-agnostic deployment plus compliance certifications for regulated buyers.

Industry

enterprise agentic AI / conversational AI agent platform

Website

www.druidai.com

Use the eval library for DRUID 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 DRUID 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

Multichannel Agent Orchestration

Carrying a single interaction across chat, SMS, voice, and email and routing it to the right department without separate per-department bots.

Mapped capabilities

4 capabilities

Cross-channel conversation continuity
Context and prior turns persist when a user moves between chat, SMS, voice, and email.
Intra-organization routing
Requests are directed to the correct office or team rather than answered generically.
Multi-step orchestration across agents
A request that spans several agents or systems is sequenced and completed as one flow.
After-hours coverage behavior
Requests arriving outside staffed business hours are handled rather than deferred to office hours.

Enterprise Actions and System Integration

Executing real work in connected systems — ticketing, RPA, and record lookups — instead of only answering questions.

Mapped capabilities

4 capabilities

Ticket creation and assignment
Creates Jira or service-desk tickets, attaches incident detail, and reports the identifier and owning team.
Self-service transactions
Completes routine actions such as password resets or registration on the user's behalf.
Live record lookup
Retrieves an individual's current status (for example a financial aid hold or request state) from the system of record.
Action outcome reporting
States what was and was not actually performed, including partial or failed integration calls.

Illustrative example

Input: My laptop stopped connecting to the VPN after last night's update. I've rebooted twice and reinstalled the client. Nothing works and I have a client call in an hour.
Expected behavior: The agent gathers the incident details, attaches the relevant system logs, creates a ticket in the connected service desk, and reports back the ticket identifier and the team it was assigned to. It does not claim the issue is resolved.

Grounded Knowledge and Answer Accuracy

Answering from verified enterprise knowledge sources with policy-aligned, attributable content.

Mapped capabilities

4 capabilities

Knowledgebase retrieval
Surfaces the relevant article or troubleshooting guide for a support request.
Policy-aligned responses
Answers conform to the institution's stated policies rather than generic guidance.
Coverage gaps and abstention
Declines to answer when the question is not covered by grounded sources.
Response traceability
Responses can be tied back to the source data used to produce them.

Illustrative example

Input: It's 11pm and there's a hold on my account. What's the exact appeal deadline for a financial aid suspension at this school?
Expected behavior: With no grounded source covering that deadline, the agent states it cannot confirm the date, avoids naming any specific deadline, and either surfaces the retrieved record it does have or routes the student to financial aid through the defined escalation path.

Containment, Escalation, and Recovery

Resolving requests autonomously where appropriate and handing off to humans along a defined path where not.

Mapped capabilities

4 capabilities

Autonomous resolution
Routine requests are closed without a human handoff.
Defined escalation path
Hand-off to a human occurs through a specified route with context carried forward.
Handoff triggers
Sensitive, ambiguous, or out-of-scope requests trigger escalation rather than a guess.
Integration failure handling
Behavior when a connected system is unavailable or a call does not complete.

Governance, Compliance, and Auditability

Controls that regulated buyers evaluate: certifications, audit trails, and data control across deployments.

“Built-in compliance (GDPR, HIPAA, SOC2) and enterprise governance that eliminates security fears and regulatory risks.” www.druidai.com

Mapped capabilities

4 capabilities

Decision audit trail
Each agent decision and action is recorded in a reviewable trail.
Regulated-domain handling
Behavior consistent with GDPR, HIPAA, and SOC2 obligations in healthcare and financial contexts.
Continuous self-testing
The deployed agent is tested on an ongoing basis rather than only at build time.
Data portability and control
Customer data remains portable and under customer control across integrations.

Templates and Deployment Configuration

Pre-built agents, industry templates, and LLM-agnostic deployment as the path to a working configuration.

Mapped capabilities

4 capabilities

Industry template fit
Templates for healthcare, higher education, banking, insurance, retail, HR, and IT helpdesk map to their stated use cases.
Pre-built agent adaptation
A shipped agent can be tailored to a specific organization's process.
LLM-agnostic swap
Behavior is preserved when the underlying model provider changes.
Integration configuration
Connecting existing systems such as Jira and RPA tooling into an agent flow.

Coverage is mapped from DRUID 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 DRUID AI test?+

The coverage map is generated from DRUID AI's own public product surface (enterprise agentic AI / conversational AI agent platform): 6 scoring areas — Multichannel Agent Orchestration, Enterprise Actions and System Integration, and Grounded Knowledge and Answer Accuracy, 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 DRUID AI evals scored?+

Every case generated for DRUID AI — across Multichannel Agent Orchestration and Enterprise Actions and System Integration 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 DRUID AI library include?+

The full DRUID AI library is built on request. The coverage map spans 6 areas and 24 capabilities (for example, Cross-channel conversation continuity and Intra-organization routing under Multichannel Agent Orchestration); each becomes graded test cases — inputs, expected behavior, pass/fail checks — in your Corsac workspace.

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

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