Private beta
Mention @meeseeks and a dedicated agent spins up for that thread — its own container, its own memory, its own job. It works until the question is answered, then checkpoints itself and disappears. Reply a week later and the same agent comes back mid-sentence.
You already have an AI bot in Slack.
It's one bot, in every channel, with one memory and no idea which conversation it's in.
Ask it something on Tuesday and it's forgotten by Wednesday.
That's not an assistant. That's a search box with a personality.
How it works
The whole lifecycle is the product. An agent is summoned into a single thread, works there, and ends itself when the work is done — without anyone remembering to turn it off.
Mention @meeseeks in any thread. A container is created for that thread alone, and the agent greets you in a few seconds.
It searches, reads and reasons with real tools. Every reply lands back in the thread, and it holds the whole context for follow-ups.
When the thread goes quiet it asks if it's still needed. No answer, and it writes a summary, saves its memory, and deletes itself.
Reply days later. A new container comes up, loads the old agent's memory, and answers like it never left.
Most bots forget when the process restarts. Here the agent's memory lives outside the machine it runs on, so crashes, restarts and deliberate shutdowns are all the same thing: a nap.
No cross-talk, no leaked context between channels, no shared scratchpad. Three threads means three isolated agents that cannot see each other.
Idle agents cost money and nobody remembers to turn them off. This one asks, waits, writes a handover note, and shuts down on its own.
Under the hood
restartPolicy: Never. Pods are meant to end.S3 and restored on the next summon. A pod is a body, not an identity.Socket Mode connection. No public ingress, no inbound webhooks, no URL to attack.IRSA, scoped to their own prefix. No static keys anywhere in the cluster.Questions
No. An agent only ever sees the one thread it was summoned into. It has no access to other channels, other threads, or your DMs.
Research and reasoning: web search, reading pages, working through a question across many turns, and summarising it all at the end. It cannot yet touch code repositories, cloud accounts, or internal systems.
Its memory is saved, and you get a written summary of the thread. Reply in that thread whenever you like and the same agent resumes with everything intact.
Join the waitlist
We're onboarding in small batches so we can actually talk to everyone. Tell us where you'd point one and we'll bump you up the list.
Where this actually is: the full lifecycle — summon, converse, hibernate, resurrect, dismiss — runs on EKS today. Agents can search and read, but they can't touch your repos or your cloud, and there's no dashboard or billing yet. We'd rather show you that than a polished demo of something that doesn't exist.