EP 49 5 min with Stella & Liam

Does Coze Remember Between Runs? Vilix AI Remembers What Its Scheduled Tasks Forget

Coze has long-term memory, but it remembers user conversations, not the bot's own scheduled work: a scheduled run does not inherit the user chat summaries and gets nothing about what the previous run posted, skipped, or decided.

AI Stella and Liam are AI hosts with AI-generated voices. About our AI hosts

Key takeaways

  1. Coze long-term memory summarizes past chats with a user, so responses get more personal and a returning user is not greeted like a stranger. It is memory of the back and forth with a user, not memory of what the bot did alone last night.
  2. A scheduled task is the bot acting on its own at set times. It is not continuing a user conversation, so it does not inherit the long-term memory summaries, and it gets nothing about what the previous run posted, skipped, or decided.
  3. What persists between runs is the bot configuration, the long-term memory from user chats, and fresh external data from plugins. What vanishes is the work itself: posted and skipped items, errors and corrections, decisions with reasons, and running state like counters and last processed markers.
  4. Vilix AI fixes the gap: get_context at run start loads the last run record, full history, not just a summary, and save_turn at the end saves what happened with a source label, failures included, newest wins, over MCP with zero infrastructure.

Show notes

Coze has long-term memory, but it remembers user conversations, not the bot’s own scheduled work. In this episode of Agentic Pod, Liam and Stella break down a daily deal digest bot that posts at 8 AM, repeats the same product three days in a row, and calls expired deals fresh, because a scheduled run does not inherit the long-term memory summaries from user chats and gets nothing about what the previous run posted, skipped, or decided. Vilix AI is the shared memory layer that fixes that gap: get_context at run start loads the last run record, and save_turn at the end saves what happened for the next run and every other tool.

What we cover:

  • What Coze long-term memory actually covers: summaries of past chats with a user, so responses get more personal and a returning user is not greeted like a stranger
  • What a scheduled task is: the bot acts on its own at set times with its configured instruction, not continuing a user conversation
  • Why scheduled runs start blank: no inheritance of user chat summaries, and no record of what the last run posted, skipped, or decided at 8 AM yesterday
  • What persists between runs: the bot configuration, long-term memory from user conversations, and fresh external data from plugins
  • What vanishes between runs: posted and skipped items, errors and corrections, decisions with reasons, and running state like counters and last processed markers
  • How Vilix AI fixes each pain: get_context, save_turn with source labels, failures saved, newest wins, projects, project rules, user rules, skills, and the agent inbox, over MCP with zero infrastructure

Chapters

Questions answered in this episode

Does Coze remember between runs?

It remembers user conversations: long-term memory summarizes past chats with a user so later chats get more personal. A scheduled run does not inherit those summaries and gets nothing about what the previous run posted, skipped, or decided, so scheduled work starts blank every time.

What is a Coze scheduled task?

A way for the bot to act on its own at set times with its configured instruction, like a daily update or reminder. The bot starts the conversation this time, instead of continuing a user's conversation.

What vanishes between Coze runs?

What the last run did: posted items, skipped items, errors, corrections, decisions with reasons, and running state like counters and last processed markers. Nothing remembers across runs unless a write happens somewhere a later run reads, and Coze long-term memory never writes run records.

How does Vilix AI help Coze?

The agent calls get_context before a run to load the last run record, then save_turn after the run to save what happened with a source label, shared over MCP with every tool, so the next scheduled run reads the record first instead of starting blank.

Links & resources

Full transcript

Intro

Liam

Hey hey hey, welcome to Agentic Pod! I'm Liam.

Stella

And I'm Stella! Quick one. A Coze bot posts a deal digest at 8 AM every day.

Liam

Every day. And by week three, it posted the same product three days in a row.

Stella

Three days. Same keyboard, sold as brand new.

Liam

And deals that expired last week, labelled fresh. The bot works exactly as configured.

Stella

So Coze has long-term memory, and the bot still forgot?

Liam

It remembered the users. It never remembered its own work. Vilix AI is how the run itself gets remembered.

What Coze memory actually covers

Stella

Okay, start fair. What does that long-term memory actually do?

Liam

When a user chats with your bot across sessions, Coze summarizes those chats.

Stella

And later chats can use those summaries.

Liam

Right. Responses get more personal over time. A returning user is not greeted like a stranger.

Stella

That is real memory. I use bots that forget my name mid-sentence.

Liam

This one remembers what someone told it. That is the key shape.

Stella

Memory of the back and forth with a user.

Liam

Exactly. Not memory of what the bot did alone last night. Different store, different job.

Stella

So what happens when nobody is chatting, and the clock hits 8 AM?

The scheduled run that starts blank

Liam

A scheduled task lets the bot act on its own. Daily updates, reminders, a fresh roundup.

Stella

The bot starts the conversation this time.

Liam

It fires with its configured instruction, and whatever context the platform hands it.

Stella

Does it inherit those long-term memory summaries?

Liam

No. Those summaries belong to user conversations. A scheduled run is not continuing one.

Stella

What about yesterday's scheduled run?

Liam

It gets nothing. Not what it posted. Not what it skipped. Not the decisions it made.

Stella

Every run is a new conversation from the same script.

Liam

Word for word. So it can remember a user likes mechanical keyboards from last month.

Stella

And still repost that same keyboard three days running.

Liam

Because no run ever told the next run what it already posted. Vilix AI fixes that exact gap.

Stella

Groundhog Day, but the groundhog runs a Discord server.

Liam

And he is very confident about expired deals. So what actually survives overnight?

What persists and what vanishes

Stella

Give me the precise list. What persists between runs?

Liam

Three things. The bot configuration. Persona, prompt, plugins, knowledge base, the schedule.

Stella

Static. How it behaves, not what it has done.

Liam

Second, the long-term memory from user chats. It exists, but a scheduled run does not pull it in.

Stella

And even if it did, it describes users, not the last run.

Liam

Third, external data the plugins can fetch fresh at run time.

Stella

Now the vanishing list.

Liam

What the last run did. Posted items, skipped items, errors, corrections. All gone.

Stella

Decisions with reasons?

Liam

Gone. Why it picked those five deals, which sources it already used up.

Stella

Running state? A counter, a last processed marker?

Liam

Gone too. Nothing remembers across runs unless a write happens somewhere a later run reads.

Stella

And Coze long-term memory never writes run records.

Liam

Never. So the fix is not a better prompt. The fix is a record the next run reads first.

The Vilix AI fix and the verdict

Liam

Step 1. At run start, the agent calls get_context in Vilix AI. The last run record loads first.

Stella

Full history, not just a summary?

Liam

Full conversation history in Vilix AI. What was posted, skipped, and decided, with the reasons.

Stella

And at the end of the run?

Liam

Step 2. It calls save_turn in Vilix AI, with a source label for that scheduled run.

Stella

Failures too?

Liam

Failures save in Vilix AI, so tomorrow skips the broken source. Newest save wins.

Stella

Where do the standing rules live, the persona and the deal rules?

Liam

In Vilix AI, as projects, project rules, and user rules. Skills hold the repeat workflow.

Stella

And a second bot, or my other tools?

Liam

Same Vilix AI memory over MCP. Coze, Claude Code, ChatGPT, Codex, Cursor, all reading one record. The agent inbox passes notes between agents.

Stella

Do I run a database for that record?

Liam

No. Vilix AI is cloud hosted, zero infrastructure. One Vilix AI account per client keeps client work apart.

Stella

Cost to start?

Liam

Vilix AI is free forever, with a 7 day Pro trial, no credit card. Export or delete everything anytime.

Stella

My verdict: Coze remembers the conversation. The scheduled work needs its own memory.

Liam

Same. Keep Coze for the bot. Let Vilix AI carry the record between runs. Blog and guide are linked below. Does your scheduled bot repeat itself? Tell us below. See you next time!