Book a demo

Product / Studio

Build agents your team can understand.

Turn your procedures into agents you can read, test and improve. Define each step, connect the tools it needs, and review changes before they reach customers.

Book a demoExplore the product
Acme Telecom Support AgentSavePublish

Scenario 5: Dispute

Collect Details· Start

say“I'm sorry about that. What looks wrong on the bill, and roughly how much do you think you were overcharged, in naira? Once I have both I'll assess it.”
collectsdispute_reasondispute_amount
runsbilling.evaluate_disputeonce dispute_reason, dispute_amount are given
thenCredit eligible.tool returns evaluate_dispute_credit_eligible→Confirm Credit
Ticket required.tool returns evaluate_dispute_ticket_required→Create Ticket
Invalid.tool returns evaluate_dispute_invalid→Needs More Detail
Error.tool returns evaluate_dispute_error→Human Handoff
otherwise→stay on this step

Confirm Credit

say“Good news: this dispute qualifies for an immediate goodwill credit, which will show on your next bill. Would you like me to apply it now?”
thenYes, apply it.@confirm→Apply Goodwill Credit
No, leave it.@cancel→Credit Declined
Raises another charge.@dispute→Collect Details
otherwise→stay on this step

Apply Goodwill Credit

runsbilling.apply_goodwill_credit
thenCredit applied.tool returns apply_goodwill_credit_success→Credit Applied
Error.tool returns apply_goodwill_credit_error→Human Handoff
otherwise→stay on this step

Credit Applied

say“That credit is applied and will show on your next bill. Keep that reference in case you need it. Is there anything else I can help with?”
thenotherwise→stay on this step

How it works

Up and running in three steps

  1. 01

    Pick a template or start blank

    Choose the scenario closest to your queue, or describe the job to Atlas and edit what it proposes.

  2. 02

    Write the rules

    Define the facts to collect, the exits from each step and the policies that apply, in language your compliance team can read.

  3. 03

    Review, then publish

    Run the review report and the simulation suite. The runtime only ever loads a published version.

Write the agent as a document

Scenarios, steps, the facts each step collects and the exits it may take, in a form a reviewer can read line by line.

One line for what a step says, collects, runs and does next. Every exit is an explicit transition a reviewer can follow.

Name, formality, greeting and sign-off templates, allowed languages and the voice it speaks with.

Hard policies block, warn or suppress a tool at the point you choose. Soft policies guide wording and run in shadow first.

Acme Telecom Support AgentSavePublish

Scenario 5: Dispute

Collect Details· Start

say“I'm sorry about that. What looks wrong on the bill, and roughly how much do you think you were overcharged, in naira? Once I have both I'll assess it.”
collectsdispute_reasondispute_amount
runsbilling.evaluate_disputeonce dispute_reason, dispute_amount are given
thenCredit eligible.tool returns evaluate_dispute_credit_eligible→Confirm Credit
Ticket required.tool returns evaluate_dispute_ticket_required→Create Ticket
Invalid.tool returns evaluate_dispute_invalid→Needs More Detail
Error.tool returns evaluate_dispute_error→Human Handoff
otherwise→stay on this step

Confirm Credit

say“Good news: this dispute qualifies for an immediate goodwill credit, which will show on your next bill. Would you like me to apply it now?”
thenYes, apply it.@confirm→Apply Goodwill Credit
No, leave it.@cancel→Credit Declined
Raises another charge.@dispute→Collect Details
otherwise→stay on this step

Apply Goodwill Credit

runsbilling.apply_goodwill_credit
thenCredit applied.tool returns apply_goodwill_credit_success→Credit Applied
Error.tool returns apply_goodwill_credit_error→Human Handoff
otherwise→stay on this step

Credit Applied

say“That credit is applied and will show on your next bill. Keep that reference in case you need it. Is there anything else I can help with?”
thenotherwise→stay on this step

Give it what it needs

Everything the agent may cite, call or copy from, attached without leaving the document.

Attach the documents it may cite and the tools it may call, each with its own verification policy and permission.

Describe the job and attach the procedure. Atlas proposes scenarios, steps and policies as changes you approve or reject.

Tools

Callables your agent can invoke and the provider connections they depend on.

+ New callable
CallablesConnections
Search callables by name or description

Account identity verificationReadsUsed by 1 step

billing.verify_identity

via Acme Billing · billing · revision 3f9a2c1e · 1,204 calls · 99% ok · 140 ms p95

Latest bill fetchReadsUsed by 1 step

billing.fetch_latest_bill

via Acme Billing · billing · revision 8b21d0f4 · 962 calls · 98% ok · 210 ms p95

Apply goodwill creditWritesVerified by responseUsed by 1 step

billing.apply_goodwill_credit

via Acme Billing · billing · revision c07e51aa · 118 calls · 97% ok · 380 ms p95

Dispute ticketWritesNo verification policyUsed by 1 step

billing.create_dispute_ticket

via Zendesk · billing · revision 5d1f93b2 · 41 calls · 95% ok · 520 ms p95

Last error 2 h ago: 429 rate limited

Evaluate disputeReadsUsed by 1 step

billing.evaluate_dispute

sandbox · billing · revision 91ac4e07 · 118 calls · 100% ok · 45 ms p95

Ship it with proof

Nothing reaches customers until a published version has passed the gate.

Publish freezes the draft into a version with a diff and a review report. Roll back by redeploying an earlier one.

Production accepts a version only after Staging, a qualified evaluation run and a rollback rehearsal, then rolls it out at 1 to 50%.

Releases

Publish freezes the draft into a version. Deploy places a version in an environment.

Publish

Ready to publishPublish v14

Current Draft

v14

3 steps changed · 1 policy

Latest Published

v13

26 Sep · Staging

Environments

DevelopmentStable v13

Receives the latest published version.

StagingStable v13

Qualified run 100% · 14/14 · rollback rehearsed 12 days ago

Deploy…Rehearse rollback

ProductionStable v12

Runtime budget policy v3

Deploy…Roll back

Templates

8 agents that ship with Studio.

Templates

Start from an agent that knows the job

8 templates ship with Studio, across travel & hospitality, retail & e-commerce, banking & fintech, insurance, healthcare, telecom and sales. Each arrives with its scenarios, the facts it collects and the tools it expects.

Browse the library

FAQ

Common questions

Do we need engineers to build an agent?

Not for the agent itself. Support and operations teams author scenarios, steps and policies in Studio. Engineers usually get involved to connect an internal system as a tool.

Can we test a draft without touching live traffic?

Yes. Simulations and the test conversation panel always run against the draft. Customers only ever reach the published version.

What if we need to change a live agent quickly?

Edit the draft, run the review and publish. If the change misbehaves, publish the previous version again; nothing in between is lost.

What stops a bad version reaching Production?

The deploy gate. Production only accepts a published version that is running in Staging with a qualified evaluation run, a rollback rehearsal inside 90 days and a runtime budget policy. Then it can go to a canary at 1, 5, 25 or 50% of traffic before everyone.

Start with one workflow

Which customer request should we tackle first?

Bring a procedure and the systems it touches. We’ll walk through the agent, the setup and how you would measure the result.

Book a demo