Moveworks Remembers the Conversation for 24 Hours. Your Scheduled Agents Still Start From Zero.

Ask your Moveworks assistant how many offices the company has in Chicago, then follow up with "where are they located," and it answers correctly. That is the context window doing its job: it remembers the ongoing conversation and resolves the reference. It feels like memory.

It is memory. But it has a clock on it, and the clock matters more than most teams realize once they start running scheduled agents next to it.

What Moveworks actually remembers

Moveworks documents its memory design openly for Agent Studio builders. The assistant uses four memory constructs when handling your request. Semantic memory holds your organization's terminology and knowledge, so "the expense policy" resolves to your company's document. Episodic memory tracks the current conversation: the questions, the answers, the decisions made in this thread. Procedure memory knows which plugins and business rules apply to what you asked. Working memory tracks what is in flight right now, including a variable-tracking system that keeps the model from mixing up business object IDs across tool calls.

Then comes the part that decides everything: the reasoner's conversational context is automatically cleared every 24 hours. Anything older than a day is removed. Moveworks says this plainly in its context-window documentation, and the reasoning is sound: stale context confuses the model, so pruning it keeps answers accurate and fast. Admins can adjust the window between 1 and 48 hours in Moveworks Setup under AI Assistant, Advanced Settings, Conversation Settings, but the default and the recommendation is 24.

The web portal adds threads: unlimited parallel conversations that keep full context, can be resumed and archived. Threads live longer, but not forever. Thirty days after the last activity, they are automatically deleted.

So the honest answer to "does Moveworks AI remember between conversations" is: yes, within the conversation, within the day, within the month. It is rolling memory, not permanent memory.

Where scheduled agents fall through the gap

Now picture the automation side of the same enterprise. A scheduled agent built in Moveworks Agent Studio, Power Automate, or ServiceNow runs every night to triage the support queue, or every Monday morning to assemble the onboarding checklist for new hires. These agents do not have a "conversation" in the sense Moveworks memory understands. They have runs. Each run starts, does its work, and ends. And every new run begins with a blank slate about the runs before it.

Three failure shapes show up again and again:

The Monday agent re-learns the same exceptions. Last week it figured out that new hires in the Berlin office need a different laptop SKU and that Dana approved the exception in a chat thread. This week it starts from zero. If Dana's thread goes quiet for a month, Moveworks deletes it, and the knowledge is gone from the system entirely.

The triage agent repeats the same dead end. Last month it escalated a category of access requests to a queue that bounced them back. Nothing recorded the rejection as a lesson, so this month it escalates the same category the same way. The agent is not learning from experience because there is nowhere for experience to live.

The context never crosses the platform boundary. Even where Moveworks built up useful conversational context, that context stays inside Moveworks. The downstream automation, the n8n workflow, the Slack bot, the email follow-up sequence, never sees it. Each tool keeps its own fragments, and the operator becomes the human memory bus, pasting context between systems.

What teams do today, and why it leaks

The standard workaround is to write run outcomes into a system of record: ticket fields, a Dataverse table, a SharePoint list, a pinned doc. Then the next run reads it back. This gets you part of the way there, but it has a failure mode that is easy to underestimate.

A raw log cannot distinguish between kinds of information. "The agent suggested a password reset" and "the user confirmed the password reset fixed it" look identical in a spreadsheet, but they are not the same kind of fact. When an agent replays its own unverified suggestions as history, it is treating guesses as ground truth, and the error compounds quietly run after run. Add schema drift, stale entries nobody cleans up, and five automations writing to five different places, and you have distributed amnesia with extra steps.

The fix: one memory your agents share

The pattern that closes this gap is simple to state: give every agent, scheduled or conversational, one shared memory it can read before acting and write to after acting. The nightly triage agent records "ticket 4821 closed with a password reset, user confirmed" when the loop closes. The onboarding agent records "Berlin hires get the X1 Carbon, exception approved by Dana, expires Nov 1." Next run, each agent checks the memory first instead of re-deriving everything from scratch, and corrections flow back in so the knowledge gets better over time instead of rotting.

Vilix AI is that layer. It is cloud-hosted with zero infrastructure for you to manage, and the same memory follows your agents across every tool and device over MCP, so the context Moveworks builds up can reach the workflows running outside it. It keeps full conversation history, not just distilled facts, which means you can always go back and check what was actually said instead of trusting a summary. The free plan is free forever, the Pro trial lasts 7 days with no credit card required, and everything is portable: export it all or delete it all, whenever you want.

The takeaway

Moveworks memory is well designed for the job it was built for: employees asking for help, day after day, in rolling conversations. The 24-hour window, the four memory constructs, the 30-day threads, all of it serves that pattern. But scheduled agents do not live in conversations. They live in runs, and they need memory that compounds across runs: what happened last night, what worked last week, what was decided last month. That is a different kind of memory, and it is worth building on purpose instead of discovering its absence one blank Monday morning at a time.