No-code Agent Studio

Build a governed AI agent in plain English — no engineering.

Agent Studio is the no-code builder for governed agents: pick skills and knowledge, set behavior and guardrails in plain English, then publish and chat-test in the same screen. Start from one of 11 role packs and ship in minutes — every reply still passes the same runtime gates and lands in the audit trail.

Book a demoRead the docs

Build → publish → chat-test · governed on the same runtime

Agent Studio
skills
Search KBCreate ticketSend email add
Publishedv4 · reliability 96
Where is order #4471?
Shipped, arriving Thursday.tools: Search KB · $0.004
studio ▸ build → publish → chat-test
The problem

Agent builders ship demos. Enterprises need governed agents.

Most no-code builders let anyone publish an agent that answers off any document, says anything, and acts with no audit trail — exactly the shadow AI your risk team is trying to stop. Cortex makes the easy path the governed path: the same Studio that takes minutes to build also compiles your behavior and guardrails into runtime policy, and refuses to publish an agent that fails its evals.

An agent below the reliability threshold cannot ship — publish is refused until it passes its evals409 RELIABILITY_BELOW_THRESHOLD
How it works

Build, publish, and chat-test — in one screen.

Agent Studio is a product surface over the registry and runtime you already trust. What you configure in the builder is what the runtime enforces — no gap between the demo and production.

  1. 01

    Build

    Pick skills and tools (with permission previews), attach knowledge sources, and set behavior — tone, length, do-not-say, escalation — all in plain English. Prompts and instructions are versioned so you can compare and roll back.

  2. 02

    Publish

    Run the test cases, then publish. Reliability gates the release: an agent below threshold is refused 409, one at or above ships as a signed, versioned definition the runtime can verify.

  3. 03

    Chat-test

    Talk to the agent in the same screen before and after publish. Every reply shows the tools it used and what it cost — and writes the same audited run record your fleet does in production.

Behavior & guardrails

Say it in plain English. The runtime enforces it for real.

Set tone and length, a do-not-say list, and escalation rules in the builder — then Cortex compiles them into policy and wraps every run in input and output guardrails. Prompt-injection attempts are blocked 451, PII is redacted on the way out, and grounded answers must carry a citation. These are not prompt suggestions the model can ignore — they run in the gated runtime.

  • Tone · length · do-not-say · escalate-to-human, all in plain English
  • Input guard: prompt-injection scan refuses poisoned input 451
  • Output guard: DLP redaction + required grounded citation before the reply is returned
See Policy-as-Code
Behavior · plain English
ToneWarm · professional
Length≤ 120 words
Do-not-sayno legal / medical advice
Escalate to humanrefund > $500 · low confidence
I/O guardrails · runtime-enforced
inputPrompt-injection scanblocked 451
outputPII / DLP redactionmasked
outputGrounded-citation checkrequired
studio ▸ guardrails compile to policy — enforced in the runtime, not the prompt
Knowledge sources
Help-center articles1,204 docs · indexed
Order ontologyOrder · Shipment · Refund
Policy handbookv12 · published
Audience rules · who sees what
Customerpublic KB only
Tier-1 agentKB + account history
Supervisor+ pricing · refund limits
knowledge ▸ audience rules resolve through ontology object & property permissions
Knowledge & audience rules

Ground answers on your data — and scope them to the viewer.

Attach help-center docs, policy handbooks, and ontology objects as governed knowledge, so the agent answers from your sources with citations instead of improvising. Audience rules decide what each viewer is allowed to see — a customer gets the public KB, a supervisor sees refund limits — and they resolve through the same object and property permissions that protect the rest of your ontology.

  • Knowledge from documents, policy handbooks, and typed ontology objects
  • Grounded answers cite their source — no ungrounded claims
  • Audience rules enforced through ontology permissions, not just the prompt
Explore the ontology
Marketplace

Start from a role pack. Ship in minutes.

Eleven governed role packs give you a working agent on day one — skills, knowledge wiring, behavior, and guardrails preconfigured for the job. Install one, point it at your data, and it inherits every runtime gate the rest of your fleet runs through.

Service

Deflect tickets, draft replies, escalate on policy.

Sales

Research accounts, draft outreach, log to CRM.

IT Desk

Triage incidents, reset access, open change tickets.

HR

Answer policy questions from the approved handbook.

Finance Ops

Match invoices, flag exceptions, route approvals.

Legal Intake

Intake matters, redact PII, draft NDAs for review.

Procurement

Compare vendors, check spend policy, raise POs.

Risk & Compliance

Screen cases, cite controls, require approval.

Field Ops

Dispatch work orders, summarize site visits.

Claims

Intake claims, triage fraud, route payouts ≥ $5k.

Tax Prep

Find deductions, ground every figure to its source.

Browse Solution Packs
Prove it — don't just claim it

The same builder, the same gates.

Try to publish a low-reliability agent and the runtime refuses it; raise the score and it ships as a signed version. Either way, every chat-test reply carries its tools and cost into the audit trail — the demo and production run the same path.

Agent Studio
skills
Search KBCreate ticketSend email add
Publishedv4 · reliability 96
Where is order #4471?
Shipped, arriving Thursday.tools: Search KB · $0.004
studio ▸ build → publish → chat-test
Publish is gated on reliability409
POST /agents/support-bot/publish   # reliability 71
  ← 409 RELIABILITY_BELOW_THRESHOLD
     fix evals, then re-run before publish

POST /agents/support-bot/publish   # reliability 96
  ← 200 { version: "v4", signed: true }
Every chat-test reply records its tools used and cost in the trust ledger
Works with the whole runtime

Studio is the front door to the governed runtime.

Everything you build in Agent Studio is the real, gated thing — so it identifies, observes, governs, and distributes like the rest of your fleet.

Agent IAM
  • Published agents become governed identities
  • Owner · risk tier · expiry
  • Allowed models & actions
Observability
  • Reliability score gates publish
  • Evals · simulation · debugger
  • Per-run quality scoring
Policy-as-Code
  • Guardrails compile to policy
  • Simulate before you ship
  • Golden tests as a gate
Solution Packs
  • Role packs install as signed packs
  • Versioned · verifiable
  • Deploy across tenants
Security & compliance

Built for the enterprise security review.

No-code does not mean no-control: every agent you build is identity-bound, guardrail-wrapped, reliability-gated, and fully audited — mapped to the frameworks your auditors already use.

SOC 2ISO 27001ISO 42001EU AI ActNIST AI RMFGDPR
Visit the Trust Center

Build a governed agent today.

Start from a role pack, set your guardrails in plain English, and chat-test in minutes — on the same runtime that keeps your whole fleet in control.

Book a demoSee the platform