Agent Studio

Design the conversation like a flowchart. Ship it like a controlled change.

Agent Studio is where your team builds every Bill Gosling agent. Set the persona and first message, lay out the steps with their branches, add the guardrails and the actions, then see what each call will cost before anything goes live. Every version is kept, so a bad change is one click from undone.

spine·voice · studio
draft v14 · 6 steps
Agent Studio · collections_outbound v14draft
Greet and disclose recordingstep 1
Verify identity2 of 3 facts
Explain balance and optionsreads account
Pay in fullpayment link
Payment plangated
Hardshiphand over
est. cost per call shown before publishSimulate
persona ✓ · guardrails 9 · tools 4 via mcp · variables bound · simulation queued · audit:chained ✓ ·
Collections agent in Studio, sample draft · rendered by spine·voice
then a named owner publishes →
SPINE · PUBLISH · collections_outbound v14live
step 01Run simulation suitescenarios and golden callsAuto · test
step 02Publish to live trafficowner approval, then rampHeld · human
✓ approved · named owner · version signed
What you build

Everything an agent says and does, in one place your operations team can read.

Scripts, policy and system access usually live in three different places. In Studio they sit on one canvas, so the people who own the call type can review exactly what the agent will do.

01

Persona and first message

Name, tone, language and the opening line, including the recording and AI disclosure. Why it matters: the opening line decides whether a customer stays on the line.

02

Visual step canvas

Tasks laid out as steps with branches for each outcome. Why it matters: supervisors and QA can check the logic without reading code.

03

Guardrails

Verify identity before disclosure, handle a wrong party, respect contact windows and never promise what policy forbids. Why it matters: compliance rules are enforced on every call, not remembered on most.

04

Actions

Look up a balance, send a payment link, book a callback or hand over to a person. Consequential actions pass through the write gate. Why it matters: the agent resolves the request instead of taking a message.

05

Variables

Customer name, balance, due date and eligible offers bound from your systems at call time. Why it matters: every conversation is specific to the customer, with no stale numbers.

06

MCP tool import

Bring in tools from any MCP server and use them as steps. Why it matters: new system access takes an import, not a project.

spine·voice // version historycollections_outbound_
Agent Studio · versionssample
VersionStateChange
v14draftHardship branch hands over earlier
v13✓ livePlan offer reads due date
v12retiredShorter greeting
v11retiredAdded callback action
est. cost per call: v14 vs v13 shownRoll back to v12
Version history, sample data · rendered by spine·voice
Cost and control

Know the cost before launch. Keep every version after it.

Two questions every leader asks about AI: what will this cost per contact, and what happens if the change goes wrong. Studio answers both on the same screen.

  • Cost estimate per callSee the expected model and voice cost of a call before you publish, and compare it with the live version. Budget conversations start from a number, not a guess.
  • Full version historyEvery edit is saved as a version with its author and its change. Auditors and QA can see what the agent said on any date.
  • Rollback in one stepRestore any earlier version if scores dip. A bad change costs minutes, not a week of complaints.
From draft to live

Studio is the first step of a release, not the last.

A version leaves Studio only through testing and a gate, so a script change never reaches your whole customer base untested.

01 · Draft

Build in Studio

Edit steps, guardrails and actions. Check the cost estimate.

02 · Test

Simulate

Synthetic callers and golden calls score the draft against your rubric.

03 · Ramp

Release behind a gate

A small share of traffic first, widened only while scores hold.

04 · Govern

Write gate

Consequential actions still stop for a policy check or a named person.

Book a demo

Bring a call script. Leave with an agent draft.

In the demo we lay out one of your call types in Studio, show its cost per call and run it past synthetic callers.