Does Apache Artemis Remember Between Runs? A Durable Message Survives in the Journal Until It Is Consumed. The Next Run Still Starts With Only Its Message.

Apache Artemis remembers a message exactly as long as delivery requires, and not one moment longer. A durable message is written to the broker's journal, survives a restart, waits through consumer downtime, and stays put until a consumer takes it and acknowledges it. Then it is deleted. That is excellent delivery memory. What Artemis never keeps is anything the consumer learned while doing the work, so an AI agent fed by Artemis queues still starts every run with only the message in front of it.

A journal built to not lose messages

Artemis, formally Apache ActiveMQ Artemis, stores persistent messages in an append-only journal: a set of files on disk, pre-created at a fixed size, that the broker fills sequentially as producers send, consumers acknowledge, and transactions open and close. Appending in order is fast, and on Linux the broker can use asynchronous IO so writes reach disk without stalling the messaging path. A garbage collection and compaction process reclaims journal files whose records are all dead, so the journal holds live state rather than growing forever.

Alongside the journal sit three more stores, each with a narrow job. Bindings record which addresses and queues exist. Large messages that exceed the journal buffer are written to their own directory. Paged messages spill to a paging directory when an address grows past its configured memory limit, so a deep backlog lives on disk instead of in heap. Persistence is on by default, and it can be switched off, at which point the broker holds everything in memory and a restart clears the board.

Notice what every one of these stores has in common. Each holds messages and routing state. None holds results. The journal can reconstruct the broker perfectly after a crash, down to which messages were delivered and which acknowledgements landed. It cannot reconstruct what any consumer concluded, because conclusions were never written there.

Consumed means deleted, and there is no replay

The defining fact about Artemis memory is the direction it runs. A queue is a hand-off, not an archive. Once every registered consumer for a durable message has received and acknowledged it, the broker removes it. Unlike a log-based system, Artemis offers no replay: a consumer that wants yesterday's message again cannot scroll back and re-read it, because yesterday's message no longer exists once its work is done.

The exceptions prove the rule, because each one still stores messages rather than knowledge. A durable subscription holds copies for a subscriber that is offline, and releases them when that subscriber returns and consumes. A last-value queue keeps only the newest message for an address, which suits prices and status values where only the latest matters. Messages that expire move to an expiry address if one is configured, and messages that fail delivery too many times move to a dead-letter address, where they wait as evidence of a problem, in their original form, for a human or a tool to inspect.

Even the journal retention feature, which can keep copies of used journal files for a configured period, is a forensic tool. It lets an operator inspect or recover broker history after an incident. It is not a feed a consumer reads, and nothing in it is offered to the next run of anything.

What this means for a scheduled AI agent

Picture the standard automation shape. A scheduled job, an order event, or a support ticket lands on an Artemis queue. A worker picks it up and hands it to an AI agent step, in n8n, Make, or a custom service. The agent does real work: it checks history, compares values, decides an exception is safe, drafts a reply in the right tone. Then it acknowledges the message, and the broker deletes the one durable record that this work ever happened.

Tomorrow the next message arrives. The agent sees a payload. It does not see that the same customer wrote last week, that the same exception was approved twice, or that the correct move last time was to escalate rather than answer. Operators usually patch this gap by hand, stuffing summaries into the next message body or into a side table the prompt must remember to read. The broker is doing its job perfectly. The gap is simply not the broker's job to close.

Delivery memory answers one question: did the message arrive? Agent work depends on a different question: what has been learned so far? Those two questions need two different systems.

Give the agent a memory that outlives the message

Vilix AI is the shared memory layer that sits beside the broker and keeps what the broker discards. Where Artemis deletes on consumption, Vilix AI keeps the full conversation history, not just extracted facts, so the reasoning behind a decision stays readable alongside the decision itself. The next run does not get a longer payload. It gets the relevant past, retrieved at the start of the run.

Vilix AI is cloud-hosted with zero infrastructure to run: no broker to cluster, no journal to size, no store to back up. Agents connect over MCP, so the same memory follows the work everywhere, from a scheduled n8n flow at night to a chat session the next morning, across tools and devices. A queue consumer, a second agent on another platform, and a human reviewing the thread all read and write one shared state instead of separate partial copies.

The working pattern is simple. The message still travels through Artemis, which remains the right tool for durable delivery. At the start of a run, the agent step pulls its context from Vilix AI. At the end of the run, it saves what it did, what it decided, and what it noticed. The broker forgets on schedule, exactly as designed, and the agent forgets nothing, exactly as required.

Starting takes minutes. The free plan is free forever, and the 7-day Pro trial needs no credit card. Data stays under full control, with export of everything or deletion of everything available at any time. Let Artemis remember the message until it is delivered. Let Vilix AI remember everything the delivery taught the agent.