Tidio's Lyro Remembers 3 Hours. Your Scheduled Agents Still Start Blank.

A shopper asks Tidio's Lyro agent about a delivery policy on Monday afternoon, comes back Wednesday morning, and has to start over. That is not a bug in your setup. Tidio's own help center documents exactly how far back Lyro can see: three hours of recent history. Inside that window, it is sharp. Outside it, the memory is gone.

If you run scheduled agents for a living, this distinction should feel familiar, because your automation agents are the same story taken further. Lyro forgets after three hours. Your nightly reporting agent forgets between every single run. Both are forms of the same architectural habit: memory scoped to the session, nothing carried forward unless you build the carry-forward yourself.

What Lyro actually remembers

Tidio is unusually specific about Lyro's memory, which makes this an easy question to answer honestly. The help center says Lyro "can see three hours into the past for the most recent history, which includes any conversation between the visitor and live agents, as well as flow and AI interactions." Three hours. That covers a single shopping session comfortably: the visitor asks about shipping, wanders to the returns page, asks about a coupon, and Lyro keeps the thread. It also covers the human handoff case, where a live agent picks up and Lyro's record of the last three hours is right there in the shared inbox.

Lyro also draws on a knowledge base you control: your FAQ pages, help center content, uploaded question-and-answer pairs, and a product catalog for stores. Tidio says the agent only answers from the data you provide, which keeps it from inventing policies or prices. When it cannot answer, it hands the conversation to a human with the full thread attached. And the Suggestions list collects the questions Lyro could not answer so you can add them to its knowledge later.

That is a real, working memory system inside one product. For the job it was designed for, answering customer questions inside a chat widget, it is enough.

What Lyro forgets

The limits show up the moment anyone leaves and comes back. A returning visitor from last week gets no recognition. A support question that references a ticket from five days ago starts cold. The knowledge base grows only when you add to it, so Lyro does not learn from outcomes on its own: it will not notice that three shoppers in a row gave up at the same answer and quietly adjust. Your analytics dashboard notices; Lyro does not.

None of this is a flaw in the product. Lyro is a customer-facing support agent with session-scoped context, and session-scoped context is what the job needs. The problem only starts when the rest of your operation assumes that memory exists everywhere. The scheduled agent that compiles Monday's Lyro conversations into a weekly report has no access to any of it. The follow-up automation that is supposed to check whether the returned item was actually refunded re-reads everything from scratch. The memory Lyro has stays inside Lyro's world.

Your scheduled agents have it worse

Put the two side by side and the contrast is sharp. Lyro remembers three hours of conversation about one visitor. Your scheduled agents remember nothing about anything. The 3 AM triage agent that reads the day's tickets and assigns priorities wakes up with no record of how it assigned them yesterday, no memory of which categories were wrong, and no idea what changed since its last run. Every run is the first run. The context window opens empty, gets filled with a re-briefing you wrote, and closes again. Anything the agent figured out along the way evaporates with it.

This is where the real cost lands. Operators deal with it by re-prompting harder, longer system prompts, more examples pasted into the briefing every cycle, and the agent still drifts because a prompt is not a memory. A prompt tells the agent what to do this run. A memory tells it what happened in the last one. Those are different jobs, and no amount of prompt engineering covers the second.

What a scheduled agent actually needs to carry forward is small and concrete: what the last run produced, which decisions are standing, what changed since, and which failures to avoid repeating. Not a dump of every token ever generated. Just the state that makes the next run smarter than the last. Three hours of chat history would be generous compared to what most automation agents get, which is zero.

One memory for the whole operation

The fix is to stop treating memory as a feature of each individual tool and start treating it as infrastructure. A hosted memory layer gives every agent, chatbot, and automation in your stack the same long-term memory through one connection. The scheduled agent that ran at 3 AM can read what the 3 AM run from yesterday concluded. The follow-up automation can see the exact state the triage agent left behind. Nothing re-briefs, nothing gets invented, and the agents share what they learn instead of each one starting blind.

This is the shape Vilix AI takes. It is a cloud-hosted memory layer, so there is no infrastructure for you to run, and the same memory follows you across every AI tool over MCP: the chatbot, the coding agent, the scheduled automation, all reading and writing the same store. It keeps full conversation history, not just extracted facts, so the real record of what happened is there to revisit. The free plan is free forever, and the 7-day Pro trial needs no credit card. Your data stays portable: export everything or delete it anytime in a portable format, and walk away with your history whenever you want.

Lyro does its job well inside its three-hour window. Your operation deserves a memory that lasts longer than a shopping session. Give your agents one memory that survives the run, and stop paying the re-briefing tax on every single cycle.

FAQ

Does Tidio's Lyro remember previous conversations? Lyro can see three hours of recent history per Tidio's help center, covering conversations with live agents, flows, and AI interactions. Beyond that window, there is no cross-session visitor memory.

Can scheduled agents share memory with Lyro? Not directly. Lyro's context stays inside Tidio's system. A shared memory layer connected over MCP is how you give your scheduled automations and your support agent access to the same record.

What should a scheduled agent remember between runs? The last run's results, standing decisions, what changed since the last run, and failures to avoid repeating. That small set of state turns each run into a continuation instead of a fresh start.

How does Vilix AI give agents memory? Vilix AI is a cloud-hosted memory layer that every tool connects to over MCP, so scheduled agents, chatbots, and coding assistants all share one memory with full conversation history. The free plan is free forever, and you can export or delete your data anytime.