Crescendo

What it does

Writes code

PartlyOnly inside its own integration workspace: Crescendo Connect turns a plain-English capability request into generated integration code and a published tool, not into application code in your repository.It reads the API documentation, plans the integration, generates the code, tests happy-path and failure-path scenarioscrescendo.ai/blog/crescendo-connect-age… ↗

Debugs and fixes

PartlyDebugging is scoped to the integrations it builds: failures in a tool run are inspected, the logic updated and rerun in the same workspace, rather than diagnosing bugs in customer software.Inspect failures, update logic, and rerun in the same workspace.crescendo.ai/blog/crescendo-connect-age… ↗

Writes tests

PartlyIt generates and runs happy-path and failure-path scenarios against each integration before the assistant may call it; these are vendor-run integration checks, not a test suite you own or edit.Run happy-path and failure-path scenarios before the assistant uses it.crescendo.ai/blog/crescendo-connect-age… ↗

Ships it

PartlyIt ships what it builds only into its own runtime: a tested integration is published as a governed tool the assistant can call in live chat and voice, with no deployment to customer infrastructure.Expose the tested capability as a governed tool for chat and voice.crescendo.ai/blog/crescendo-connect-age… ↗

Done-for-you service

YesThis is software plus people: Crescendo deploys its own team to calibrate the platform, and its PartnerHero affiliates in seven countries supply human support agents and QA.We deploy a team that keeps the platform calibrated to your business and owns the outcome.crescendo.ai ↗

How much it does unattended

Runs autonomously

YesThe Concierge agent is described as closing issues end to end — status lookups, account updates, order changes — and handing off with full context only when it cannot resolve.Agentic infrastructure means that Concierge understands customer intent, takes action, and closes issues without a human in the loop.crescendo.ai/agents/concierge ↗

Multi-agent

YesSeven named agents divide the operation — Concierge, Quality, Agent Assist, Applied Insights, Optimization, Integration and Knowledge — and feed each other in a documented improvement loop.A team of autonomous CX agents that work 24/7 so your team can focus on the high-judgment work.crescendo.ai ↗

Agent permissions

YesA named Safe Action Framework splits autonomous actions from ones needing human approval, and published integration tools carry typed inputs, permissions and review before the agent may call them.Tenant-isolated, audit-logged, and governed by a Safe Action Framework that determines what AI can do autonomously and what requires human approval.crescendo.ai ↗

Which models it runs on

Claude

NoAnthropic appears nowhere on the site, and the binding sub-processor list that authorises every third party touching customer data names OpenAI, Google Gemini/Vertex, Groq, Azure and Amazon Nova Sonic only.Third-party providers Crescendo uses to process personal data; the Use column clarifies whether each supports the Platform, Admin, or Both.crescendo.ai/subprocessors ↗

GPT

YesOpenAI is a listed sub-processor and Crescendo announced GPT-5 across CX Assistant, Voice of Customer and CX Insights, available to all customers rather than gated to a tier.GPT-5 is available in all Crescendo environments.crescendo.ai/news/crescendo-adds-gpt-5 ↗

Model choice

NoThe vendor picks: requests are fanned out across several LLM providers in parallel and the fastest valid response wins, with no published control for a buyer to pin a model.Crescendo’s multi-provider architecture routes every AI request across multiple large language model (LLM) providers in parallel, and automatically…crescendo.ai/news/multi-provider-archit… ↗

Not established: Open models

Where you use it

In a browser

YesEverything runs in the browser: an embedded chat widget on the customer's site and app, plus a hosted dashboard where conversations, routing and analytics live.Manage all support conversations in one streamlined dashboard.crescendo.ai/solutions/features ↗

Whose machine it runs on

Self-hosted

NoDelivered as software-as-a-service only; the platform runs on Google Cloud Platform with US sub-processors, and no on-premise, VPC or bring-your-own-infrastructure option is offered on any tier.Crescendo’s software-as-a-service platform provided to Customer as part of the Service(s), which is distinct from End User Support Service and any…crescendo.ai/policies/platform-terms ↗

Open source

NoProprietary throughout: no repository, licence or source release is published anywhere on the site, and the platform terms forbid reverse engineering or adapting the service.reverse engineer, modify, adapt, or hack the Service, or otherwise attempt to gain unauthorized access to the Service or its related systems or…crescendo.ai/policies/platform-terms ↗

What it costs to run

How it meters

PartlyBilling is per resolution — $1.25 for AI-handled, $2.25 for the AI-plus-human model — but the fixed monthly fee charged alongside it is never quantified, so only half the unit is public.It charges a flat rate of $1.25 per resolution and a fixed monthly fee.crescendo.ai/blog/how-much-do-chatbots-… ↗

Free tier

NoNo $0 plan, self-serve signup or trial is published; the single route in is a booked demo, and the only free items are onboarding and the guarantee credits.Book a personalized demo with us. We’ll show you how our technology, and our agents, can elevate and evolve your customer experience.crescendo.ai/contact ↗

API access

PartlyDevelopers can drive the embedded assistant from their own page through widget functions that pass context, but no public REST API or developer documentation for the platform is published.Web developers can now invoke functions to send contextual information or set pre-filled fields in the CX Assistant widget.crescendo.ai/blog/crescendo-technology-… ↗

MCP server

PartlyMCP runs the other way round: Crescendo's assistants act as clients against third-party MCP servers such as Shopify's, and no Crescendo MCP server is published for outside agents to query.Crescendo adds additional support for Model Context Protocol (MCP) providing new capabilities that allow Assistants to interface more effectively to…crescendo.ai/blog/crescendo-ai-suite-20… ↗

Bring your own key

NoInference is Crescendo's to arrange: the contract defines LLM providers as entities it retains, and the only customer lever is objecting to a named sub-processor within thirty days.any entity retained or used by Crescendo to provide certain AI features, including but not limited to chatbot functionalitycrescendo.ai/policies/platform-terms ↗

Buying it for a team

A company can buy it

YesSold to organisations only, under a master agreement and executed Service Orders, with the platform licensed for the customer's internal business operations and its end users.make the Platform available to Customer for use by Authorized Users solely for the internal business operations of Customer and End-Users via…crescendo.ai/policies/platform-terms ↗

Seat model

NoNot a per-seat product: Crescendo prices the resolution and says so explicitly against seat-selling rivals, so no seat floor, ceiling or per-agent rate is published.While others sell seats, licenses, or hype, we tie our success to business outcomes.crescendo.ai/news/multi-provider-archit… ↗

Pooled budget

PartlyBuyers hold one contracted conversation volume rather than per-seat balances, and low-CSAT conversations are excluded from it; no allocation or sub-budget controls are described.under our Quality Guarantee, any conversation scoring 2 or lower on Predictive CSAT doesn't count toward your contracted volume at all.crescendo.ai/crescendo-vs-fin ↗

Admin controls

YesAdmins constrain both people and agent: role-based permissions govern who sees what, and the Safe Action Framework decides which actions the AI may take unattended.Control who can see and do what with flexible access roles.crescendo.ai/solutions/features ↗

Audit log

YesAccess logging and real-time activity tracking are listed platform features, and the homepage describes every tenant as audit-logged; retention windows are configurable.Track who accessed what and when, for added visibility and security.crescendo.ai/solutions/features ↗

Not established: Single sign-on

What happens to your code

Opt out of training

PartlyThe DPA bars use of customer personal data beyond delivering the service, but there is no published model-training opt-out, and de-identified derivatives of your data become Crescendo's property.Crescendo will not: (i) retain, use, or disclose Customer Personal Data for any purpose other than for the specific purpose of performing the…crescendo.ai/policies/data-processing-a… ↗

Data residency

PartlyNo region choice is offered: every listed sub-processor sits in the United States and the DPA reserves global transfer, with EU and UK transfers covered by Standard Contractual Clauses.Crescendo may transfer and Process Customer Personal Data on a global basis as necessary to provide the Service(s).crescendo.ai/policies/data-processing-a… ↗

Getting out

YesReports and insights export to CSV, Excel, image or PDF in-product, and on termination the customer instructs deletion or return of all personal data; no credits exist to expire.Following the termination or expiration of the Agreement, Customer shall provide written instructions to Crescendo to delete or return all Customer…crescendo.ai/policies/data-processing-a… ↗

Certifications

YesSOC 2 Type II, ISO 27001 and HIPAA attestation with GDPR- and CCPA-aligned controls; certificates and the DPA are requestable from the trust portal, and a BAA is available.SOC 2 Type II and HIPAA attested, with GDPR- and CCPA-aligned controls, ISO 27001 certification, and automated PII and PHI redaction during exportcrescendo.ai/news/multi-provider-archit… ↗