Find the right answer
Search approved documents, retrieve policy sections, summarise controlled knowledge and cite source material.
Platform - Agent Builder
Build and host enterprise agents without turning every idea into a custom app.
Agent Builder is the product surface for creating governed business agents: define the purpose, teach the agent with approved knowledge, equip it with reviewed tools, test behaviour, request approval, publish to the right audience and host the agent as a named enterprise chat experience.
agent drafttravel-expense-compliance-assistantready for review
OwnerFinance OperationsBusiness-managed agent
AudienceEmployeesScoped launch group
Knowledge4 packsPolicy, FAQ, receipts, exceptions
Tools3 approvedLookup, calculator, handoff
What the product does
Agent Builder is where enterprise agents are created, hosted and improved.
AI Enablement is the broader programme. Agent Builder is the product teams use to turn high-volume business work into named agents that employees can launch. It separates the creator experience from the admin configuration, so end users get a focused agent chat while owners and reviewers keep control of knowledge, tools, approvals and version history.
- Create: define purpose, audience, owner, instructions, welcome message and suggested questions.
- Teach: attach approved documents, FAQs, policy packs, examples and controlled knowledge sources.
- Equip: select reviewed tools, MCP capabilities, workflow actions and model routes.
- Approve: route sensitive knowledge, powerful tools or wider publication through review.
- Host: launch a named chat session for enterprise users and keep the agent measurable over time.
01Creator workspaceBusiness owners shape the agent without editing low-level runtime configuration.
02Review workflowOwners test answers, approvals capture launch decisions, and changes stay versioned.
03Hosted agent sessionUsers launch the named agent into a clean chat experience focused on the task.
04Operating evidenceUsage, cost, tool calls and policy outcomes are available for improvement and review.
Builder journey
From recurring question to governed agent in one product flow.
A creator should not start with identity architecture or system prompts. They should start with a business problem and move through a guided path that makes the agent useful before it becomes broadly available.
Step 1DefineName the agent, describe the job, choose the owner group and set the intended audience.
Step 2TeachAdd instructions, welcome prompts, examples, policies, FAQs and controlled knowledge packs.
Step 3EquipSelect approved model routes, tools, toolsets, MCP capabilities and workflow handoffs.
Step 4TestPreview the chat, run sample questions, refine behaviour and collect owner feedback.
Step 5PublishRequest approval where required and launch a named agent session to the intended users.
Example build
Travel & Expense Compliance Assistant.
The finance team is tired of answering the same policy questions: can I claim a client dinner, what receipt is required, is an upgraded hotel reimbursable, what happens if my train was delayed, how do I handle mileage or foreign currency?
Instead of sending employees to public chat tools or asking IT for a bespoke application, Finance creates a hosted agent in AI Warden. The agent answers from approved policy, cites the relevant rule when possible, uses reviewed tools for calculation and escalates uncertain cases to Finance.
- Purpose: answer travel and expense questions using approved company policy.
- Audience: employees, with Finance Operations as owner and escalation path.
- Knowledge: travel policy, reimbursement rules, receipt examples and FAQ.
- Tools: policy lookup, allowance calculator and finance ticket handoff.
- Launch goal: reduce repetitive questions while routing exceptions to the right team.
live previewTravel & Expense Compliance Assistant
EmployeeCan I claim dinner after a delayed train?
Travel & Expense Compliance AssistantUsually yes if the delay exceeds the policy threshold and you attach the receipt. I found this in section 4.2. If this was client-related, I can open a Finance handoff ticket.
Grounded answer · approved handoff · policy citation requested
Teach the agent
Instructions, knowledge and examples make the agent feel like the department built it.
Agent Builder gives non-technical owners a structured way to describe how the agent should behave. They can define tone, boundaries, escalation criteria and example conversations, then attach approved content that grounds the answers.
- Agent instructions: what the agent does, what it should avoid, when to escalate and how to cite policy.
- Welcome experience: opening message, suggested questions and clear explanation of what the agent can help with.
- Knowledge packs: policies, FAQs, playbooks, controlled documents, exports or retrieval-backed sources.
- Example conversations: good answers, bad answers, escalation cases and edge cases for testing.
agent: travel-expense-compliance-assistant owner: finance operations goal: answer employee travel and expense questions instruction: Answer from approved Finance documents. Cite the relevant policy section when possible. Ask follow-up questions before eligibility guidance. Escalate ambiguous exception requests to Finance. suggested_questions: - Can I claim dinner after a delayed train? - What receipt do I need for a hotel? - Is mileage reimbursable for client travel?
Tools and capabilities
Agents become valuable when they can use the right approved tools.
Agent Builder lets owners attach reviewed capabilities as part of the build journey. The builder stays simple for business teams, while platform owners control what each tool can do and who can use it.
Do useful work
Calculate allowances, check eligibility, create service tickets, hand off exceptions or prepare structured forms.
Connect governed systems
Use approved MCP servers and internal API tools while keeping request and response inspection in the gateway path.
Package repeatable capabilities
Bundle related tools, instructions and boundaries so teams select a reviewed pack instead of assembling plugins one by one.
Use approved models
Select governed model aliases from the LLM Gateway so each agent has the right route, budget and policy envelope.
React to events
Warn users, disable selected tools, end sessions, pause agents or escalate to reviewers after policy or DLP outcomes.
Test, review and publish
Experiment safely before an agent reaches enterprise users.
Business teams can test draft agents with sample conversations and refine instructions before launch. When the agent reaches a wider audience, uses sensitive knowledge, gains powerful tools or has a higher budget, approval rules route the change to the right owners.
- Preview chat as the named agent before publishing.
- Run example questions against attached knowledge and tools.
- Collect feedback from business owners and reviewers.
- Require approval for sensitive data, external connectivity, write actions, higher budgets or wider publication.
- Keep version history so agents can be changed safely and rolled back when needed.
BehaviourPassAnswers cite policy and escalate exceptions.
KnowledgeApprovedFinance policy pack version 2026-Q3.
ToolsScopedLookup, calculator and handoff only.
LaunchReviewFinance approval required before publishing.
Host and operate
Published agents become measurable enterprise services.
Once approved, agents can be launched by the intended users in a focused chat session. Owners can see adoption, common questions, tool usage, cost, policy outcomes and improvement opportunities without exposing end users to system prompts or configuration details.
What the user sees
- A named agent, not a generic platform persona.
- A clean chat session focused on the task.
- Suggested prompts and clear boundaries.
- Answers grounded in approved content and tools.
- Escalation when the agent should not decide.
What the owner manages
- Agent configuration, knowledge, tools and launch scope.
- Review workflow, version history and publication status.
- Usage, cost, policy events and DLP outcomes.
- Feedback loops for unclear policy or missing knowledge.
- Retirement or pause when the agent is no longer needed.
Business-built, enterprise-hosted
Give every department a safe path from idea to hosted AI agent.
Agent Builder is the product for turning internal expertise into governed AI services with approved knowledge, tools, workflows, launch controls and named chat sessions.