EP 51 5 min with Stella & Liam

Does a Slack AI Agent Remember Between Runs? Vilix AI Remembers the Channel, Not Just the Thread

Your Slack AI agent is brilliant inside one thread and blank in the next one: a custom agent reads only the thread it was summoned in, so a new thread starts blank and old decisions get invented.

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

Key takeaways

  1. Slack AI search is retrieval over stored messages, channels, files and canvases, not memory of what the agent did or decided. The agent itself keeps nothing, and a custom agent reads only the thread it was summoned in.
  2. A new thread or DM starts blank: the 7 AM digest cannot say which tickets are still open, and the on-call agent never saw the first incident's fix, because both live in other threads.
  3. The worst failure is the invented answer: an agent that cannot remember a decision invents one, with full confidence, inside a thread the team trusts.
  4. Search is not memory: retrieval waits for the right question, while memory carries context forward on its own. The DIY fix turns your memory into a second codebase with its own schema and failure modes.
  5. Vilix AI fixes the gap: get_context at thread start loads the decision and its reasoning, and save_turn at the end saves the thread with a source label, failures included, newest wins, over MCP with zero infrastructure.

Show notes

Your Slack AI agent is brilliant inside one thread and blank in the next one. In this episode of Agentic Pod, Liam and Stella break down the thread boundary: a custom Slack agent reads only the thread it was summoned in, so a new thread asking what the team decided about the Acme refund last week gets silence, then an invented answer. Vilix AI is the shared memory layer under the agent: get_context at thread start loads the decision and its reasoning, and save_turn at the end saves the thread for the next run and every other tool.

What we cover:

  • What Slack AI search actually gives you: retrieval over stored messages, channels, files and canvases, not memory of what the agent did or decided
  • How custom Slack agents work: the summoned thread is the whole working memory, and a new thread or DM starts blank
  • The daily burns: the 7 AM digest that cannot say which tickets are still open, and the on-call agent that never saw the first incident’s fix
  • The worst failure: an agent that cannot remember a decision invents one, with full confidence, inside a thread the team trusts
  • Why search is not memory, and why the DIY fix turns your memory into a second codebase with its own schema and failure modes
  • 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 a Slack AI agent remember between runs?

Inside one thread, yes: the thread is its working memory. A new thread, a new DM, or a scheduled run in a fresh thread starts blank. Slack AI search can retrieve stored messages, but that is retrieval over what people wrote, not memory of what the agent did or decided.

What does Slack AI search actually give an agent?

Search over the messages Slack stores, plus files and canvases, inside what the user is allowed to see. It answers when you ask the right question. It does not carry context forward on its own, and it holds nothing about the agent's own past decisions, open items, or failures.

Why does a Slack agent invent answers?

When an agent cannot remember a decision, it often fills the gap with a plausible one, stated with full confidence. The fix is a real record the agent reads first: the decision and its reasoning, saved where the next thread can load it.

How does Vilix AI help a Slack agent?

The agent calls get_context at thread start to load the full history that matters, then save_turn at the end to save the thread with a source label, shared over MCP with every tool, so the next thread, the digest, and the on-call agent read 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! Picture this. Your Slack agent nails a support thread.

Liam

It reads every reply. It cites the fix from Tuesday. Brilliant.

Stella

Then you open a new thread. What did we decide about the Acme refund last week?

Liam

Silence. Then a confident answer your team never made.

Stella

It invented a refund decision?

Liam

Word for word, with total confidence. The thread ended, so the memory ended. Vilix AI is how the agent remembers the whole channel.

What Slack search actually gives you

Stella

Start fair. Slack AI itself can search months of chat, right?

Liam

Right. In early 2026, Slack rebuilt Slackbot as a context aware agent.

Stella

It pulls from your channels, your files, your canvases.

Liam

And it drafts briefings in your voice, inside what you are allowed to see.

Stella

That sounds like memory.

Liam

It is retrieval. Slack stores the messages, so the agent can search them.

Stella

Like me with the search bar, just faster.

Liam

Exactly. The agent itself keeps nothing. Not what it did. Not what it decided.

Stella

Close the chat, open a new one, and it starts from search alone.

Liam

And a custom agent you build yourself gets even less. So what does it actually see?

The thread boundary in production

Liam

A custom Slack agent reads the thread it was summoned in. That thread is its whole working memory.

Stella

Great at summarize this thread.

Liam

Useless at what did the last three Acme threads say. A new thread is a blank slate.

Stella

A DM about another topic?

Liam

Blank there too. Now watch the daily burns. The digest agent runs at 7 AM.

Stella

Summarizing overnight tickets in a fresh thread.

Liam

Yesterday's digest is not in its memory. It cannot say which tickets are still open.

Stella

The on-call assistant next. A second incident, same service, two days later.

Liam

It has no idea what fixed the first one. The weird database flag, the runbook step, all in another thread.

Stella

Which is to say, nowhere.

Liam

Nowhere with great formatting. And the worst failure is the invented answer.

Stella

The Acme refund that never happened.

Liam

An agent that cannot remember a decision often invents one instead. Same confidence, wrong facts, inside a thread your team trusts. Vilix AI closes that gap, and we will get to how.

Stella

Hold on. Why can search not just fix all of this?

Search is not memory

Liam

Because retrieval waits for the right question. Memory carries the context forward on its own.

Stella

Give me the test.

Liam

An agent with memory checks its own state before it answers.

Stella

It spots the customer who complained twice.

Liam

It follows up on the task it promised yesterday. Search never volunteers any of that.

Stella

So builders wire it up themselves?

Liam

They can. Write every decision into a database at the end of each run. Read it back at the start of the next.

Stella

That works, no?

Liam

It works. And now your memory is a second codebase. Schema, versioning, failure modes, maintained forever.

Stella

Every agent team becomes a database team.

Liam

Every single one. Which raises the real question. What if the memory layer already exists?

The Vilix AI fix and the verdict

Liam

Step one. At thread start, the agent calls get_context in Vilix AI.

Stella

What loads?

Liam

The full history that matters. The Acme decision, with the reasoning behind it. Full conversations in Vilix AI, not just facts.

Stella

Step two, at the end?

Liam

It calls save_turn in Vilix AI, with a source label for that Slack thread and that run.

Stella

So the digest at 7 AM reads yesterday's record first.

Liam

Exactly. Open tickets carry over. The on-call agent reads the first incident's fix. Failures save in Vilix AI too, so the next run skips the broken path.

Stella

And when two records disagree?

Liam

Newest save wins in Vilix AI. Standing rules live as projects, project rules, and user rules. Repeat workflows live as skills.

Stella

My other tools, Claude Code, Cursor, my phone?

Liam

Same Vilix AI memory over MCP. One memory under the Slack bot, the scheduled runs, and every tool you open. The agent inbox passes notes between agents.

Stella

Do I run servers for that?

Liam

No. Vilix AI is cloud hosted, zero infrastructure. Agencies give each client one Vilix AI account, so client memory stays apart.

Stella

Cost to try it?

Liam

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

Stella

My verdict: the thread is a room. Your agent needs the whole building.

Liam

Same. Keep Slack for the conversation. Let Vilix AI carry the channel memory between threads. Blog and guide are linked below. Does your Slack agent remember the last thread? Tell us below. See you next time!