All systems operational
SOC 2 Type II · Trust center →
AI Workflows

A growing library of agentic workflows, pre-built for support.

Pick one. Run it on real data within a day. Each workflow is a coordinated set of agents — planner, retriever, executor, verifier, writer — with public evals, audit, and approval policies baked in.

Architecture

Five agents. One outcome.

01 · Planner
Decompose the issue

Classify severity, blast radius, downstream effects.

02 · Retriever
Ground in context

Hybrid BM25 + dense + reranker over docs & telemetry.

03 · Executor
Take action

Tool-use across 200+ systems with idempotent guards.

04 · Verifier
Score & gate

Independent model checks factuality, safety, policy.

05 · Writer
Close the loop

Customer message, internal note, audit trail entry.

Workflow library

Forty-two workflows, shipping more weekly.

billing.recover

Failed payment recovery

Detect Stripe 402. Retry with updated method. Reach out if it persists.

Live
integration.refresh

Auto OAuth re-auth

Detect expired token. Initiate re-auth flow. Notify customer if action needed.

Live
activation.nudge

Stuck-in-onboarding

Identify drop-off at step N. Personalize a 60-second fix and ship it.

Live
infra.diagnose

Webhook diagnostics

Investigate failing webhook deliveries and propose a fix to the customer.

Live
account.hygiene

Account hygiene

Daily sweep for dormant invites, stale seats, broken integrations.

Live
ticket.triage

Inbound triage

Classify, route, draft response, attach related runs, hand off if needed.

Live
refund.guard

Policy-bounded refunds

Honor refund within policy. Escalate above threshold with full context.

Live
churn.signal

Churn intercept

Detect risk signal in usage curves. Coordinate with CSM via Slack.

New
voice.deflect

Voice-first deflection

Sub-300ms voice agent for tier-1 phone deflection with handoff to humans.

Beta
Build your own

Describe the outcome. Get a workflow.

Quellra's natural-language workflow builder writes the DAG, picks the tools, generates evals, and ships it behind your approval policies. You stay in control; the agents do the typing.

Try the builder →
quellra · workflow.compileauto-saved
// describe the outcome you want
workflow "recover_failed_payments" {
  on  stripe.charge.failed
  when customer.plan in ["growth", "enterprise"]

  retrieve stripe.payment_methods(customer)
  retrieve docs.help["failed_payments"]

  try stripe.charge.retry(payment_method = latest_valid)
  otherwise
    notify customer.email
      tone   = "calm"
      include = link.update_payment_method

  verify outcome.resolved == true
  log audit.append(workflow, trace)
}

Your first workflow runs in production within 14 days.