Tidio

Two routes out, one bounded and one general. Actions call external services over their APIs for third-party data and write back to them, with native Shopify support and coexistence alongside an existing helpdesk such as Zendesk.

Above that, Tidio documents Model Context Protocol support: connect Lyro once to a CRM, billing platform or database and it can use every tool that system exposes, with named servers including Shopify, Guru and Stripe and the option to ship your own for product search, bookings or order status. Tidio states every call is scoped, authenticated and auditable. That makes breadth bounded by what the customer connects rather than by a catalog.

Sourcetidio.com/ai-agent/lyro-actionsread 2026-09-05

Lyro executes rather than answers. Actions let the agent retrieve and modify information in external systems dynamically mid-conversation over their APIs, checking order status and shipment details, managing address changes, updating databases, creating tickets and scheduling appointments, with an API call or a sequence of calls triggered when Lyro determines a customer needs external information.

Tidio positions this explicitly against rigid scripted flows, and the extent of what is possible is bounded by the external system's API rather than by a fixed action list. The Smart Actions and Simple Actions distinction is now legacy and survives only on a small number of older projects; the current product presents a single Actions tool.

Sourcehelp.tidio.com/hc/en-us/articles/20344365583132-lyro-actionsread 2026-09-05

A maintained retrieval structure the operator curates and scopes. A unified knowledge base feeds both customer-facing automation and agent assistance from one source, adding a website URL automatically extracts and structures the content into searchable knowledge, and operators control which sources serve customer responses versus internal assistance.

Tidio documents that Lyro normally depends on the knowledge uploaded into it and does not improvise outside it, with Actions existing precisely to cover cases that need third-party data the knowledge base cannot supply, which is an unusually clean statement of where grounding ends.

Sourcehelp.tidio.com/hc/en-us/articles/20344365583132-lyro-actionsread 2026-09-05

Control sits with the operator at configuration time and at the handoff. Guidance lets operators author behavioral instruction scoped to the Lyro agent covering tone, style and escalation rules, operators retain complete control over which Actions Lyro can perform and which systems it can access, individual Actions can be tested before activation and enabled or disabled manually, and unresolved queries hand off instantly to a human.

The vendor states the limit plainly: Tidio documents that Lyro does not validate or roll back integrations, instructs customers to test in a safe environment, and states that unintended changes to external systems are irreversible by Tidio, with a further note that the customer's security team should evaluate the risk of data exposure. That is a real boundary on the oversight the platform provides, and it is the customer's configuration rather than a Tidio gate that stops a bad write.

Sourcehelp.tidio.com/hc/en-us/articles/20344365583132-lyro-actionsread 2026-09-05

Attestation plus named customer-facing controls, published through a live trust portal covering Tidio services including the Lyro AI Agent. An independent third party completed a SOC 2 Type II audit covering Security, Availability and Confidentiality, and the report, an annual penetration test confirmation letter and a HECVAT-Lite are all available on request through the portal. Tidio also cites GDPR, CCPA and the EU voluntary AI Pact, and its Responsible Disclosure Policy offers safe harbor for good-faith security testing.

Data is protected by TLS 1.2+ in transit and AES-256 at rest, accounts support two-factor authentication via TOTP with backup codes, a DPA is folded into the terms, and a version-controlled subprocessor list is published and updated at least thirty days before any change, which is a strong disclosure commitment. Tidio states plainly that it is not HIPAA compliant and will not sign a BAA.

Sourcetrust.tidio.comread 2026-09-05

Call-level auditability is claimed and run-level reconstruction is not described. Tidio states that every Action and MCP call is scoped, authenticated and auditable, the Actions list reports an execution count per action over the trailing thirty days, and advanced analytics ship from the Growth tier. No agent decision trace, reasoning log or configuration audit trail is documented, and auditable is asserted without a described surface, so a buyer cannot see where the record lives or what it contains.

Sourcetidio.com/ai-agent/lyro-actionsread 2026-09-05

In-session only, stated by the vendor rather than inferred. Tidio's security documentation states that Lyro works in session only, so no state persists between conversations, and the Actions and MCP mechanisms retrieve CRM notes, past tickets and subscription state from external systems at request time, which is live retrieval against the customer's own systems rather than agent memory. Reading a record the customer's CRM holds is not a memory layer, whoever writes it. Here the absence is documented by the vendor rather than concluded from silence.

Sourcehelp.tidio.com/hc/en-us/articles/20344365583132-lyro-actionsread 2026-09-05

A residency commitment exists and no customer selection does. Tidio documents that all customer data is stored on servers located in European Economic Area member countries, hosted on AWS in ISO-certified EU data centers, which is a stated location rather than a silence.

It is a single fixed region with no customer choice of where data sits, and no self-hosted, private-cloud or single-tenant option is published. This bears on what the commitment is worth: the parent entity Tidio LLC is US incorporated, so EU storage does not by itself remove US legal reach, and the trust portal covers Tidio services including Lyro under one policy.

Sourcetrust.tidio.comread 2026-09-05

A browsable template library ships in the product, templates can be added alongside custom Actions, and Actions can be cloned and exported for reuse across projects, so the customer adopts packaged assets by selection rather than authoring each one. Tidio also drops in ready MCP servers including Shopify, Guru and Stripe as adoptable units. The unit adopted is an Action or a server rather than a whole agent, and no catalog of prebuilt agents exists.

Sourcehelp.tidio.com/hc/en-us/articles/20344365583132-lyro-actionsread 2026-09-05

Lyro runs on the live chat widget by default and is switched on per channel for Facebook Messenger, Instagram, WhatsApp and incoming email, with connected mailboxes listed automatically and tickets answered from the same inbox; the multichannel product connects email, WhatsApp, Messenger, Instagram and live chat into one place, and Lyro audiences can branch on which channel a visitor arrived through. Voice and SMS are not among the channels.

Sourcehelp.tidio.com/hc/en-us/articles/9003475527196-lyro-the-conversational-ai-agentread 2026-09-05

Lyro's documented configuration surface covers channels, audiences, knowledge sources and behavioral guidance, with no model selection, routing or bring-your-own-model control anywhere in it, and the underlying provider is not named.

Sourcehelp.tidio.com/hc/en-us/articles/9003475527196-lyro-the-conversational-ai-agentread 2026-09-05

Callable from outside, the platform extends in more than one direction: documented Model Context Protocol support connects Lyro to a system once so it can use every tool that system exposes, with customers able to ship their own MCP servers for product search, bookings or order status rather than only consuming Tidio's; an OpenAPI surface is available on the Plus tier; Actions are authored directly against external APIs; and Actions can be exported for portability between projects. Tidio notes the Actions tooling is intended for developers, which is a statement about audience rather than availability.

Sourcetidio.com/ai-agent/lyro-actionsread 2026-09-05

A component-level gate is documented; an agent-level scored evaluation is not. Individual Actions can be tested inside the editor before deploying to a live site, drafts are flagged in the Actions list, and Tidio instructs customers to test in a safe environment because it does not validate or roll back integrations.

No scored evaluation, scenario set or readable comparable verdict on a proposed agent change is documented, so a team can exercise an action but not compare two versions of an agent. Secondary guides citing Tidio's help center describe a Playground sandbox for testing Lyro on chat and email before going live, with the ability to add a Q and A pair on the spot when it answers wrong. Neither that sandbox nor the action test produces a score.

Sourcehelp.tidio.com/hc/en-us/articles/20344365583132-lyro-actionsread 2026-09-05

No browser control, hosted session or computer use appears anywhere in Tidio's product or help documentation. Lyro works programmatically, acting through Actions and Model Context Protocol connections against external APIs, with every call scoped and authenticated, and it reaches customers through the widget and its messaging channels rather than by operating an interface.

Sourcetidio.com/ai-agent/lyro-actionsread 2026-09-05