Skip to content
engineer9 docs

How it works

engineer9 is a teammate that can automate. It lives in your Slack, learns who does what and how your systems behave, and steps in when it can genuinely help. Everything below is how that works.

A shared brain for the team

Section titled “A shared brain for the team”

Most AI agents are single-player: you prompt, they answer, and the context dies with the session. Teams don’t work that way. The knowledge that keeps a team running is who owns which service, why a decision was made, and how a past incident was resolved. That knowledge is spread across people and tools, and most of it is never written down.

engineer9 captures that knowledge as it flows through Slack and keeps it. The longer it’s in your workspace, the better its answers get, because it’s answering from your team’s accumulated context rather than a generic model.

Memory

Section titled “Memory”

Everything engineer9 knows is its Memory: team decisions, who owns what, how things were fixed before. It reads from Memory before answering and writes to it when it learns something new.

Two behaviors make Memory more than a database:

  • Surface: when you ask a question, engineer9 retrieves the relevant memory and presents it in context rather than as a raw dump.
  • Reflect: engineer9 revisits what it already knows and comes back with new insight on top of it. It connects what it has learned across many conversations instead of recalling a single stored note.

Ownership detection

Section titled “Ownership detection”

From activity in the channels you’ve given it access to, engineer9 builds a Roster: a map of who owns which component or area of expertise. It uses this to route to the right person. When something needs a specific owner’s input or approval, it pings them rather than the whole channel.

The Roster builds over a few days of listening. An admin can correct it at any time (see Channels & access).

Acting on a heartbeat

Section titled “Acting on a heartbeat”

engineer9 is reactive by default and responds when you @mention it. Because it’s a teammate, it also acts on its own on a scheduled heartbeat: at each tick it looks at recent channel activity and decides whether to step in. It initiates a message only when one of these is true:

  • An unanswered question it knows the answer to, where no one has replied for at least 30 minutes.
  • A stale incident: an unresolved signal older than 2 hours with no resolution noted, where it asks whether the issue was resolved and offers to write a playbook.
  • An unknown system or person it has no record of, which it asks about once.

It posts at most one message per tick, never starts a top-level message when there’s a thread to reply in, and never follows up the same thread more than once a day.

Knowing when to stay quiet

Section titled “Knowing when to stay quiet”

The most important part of engineer9’s agency is restraint. If it isn’t confident it knows the answer, it does nothing. No guess, no noise. If someone has already handled a thread, it stays out of it.

The same honesty applies when you ask it directly: if it doesn’t know, it tells you it doesn’t know rather than fabricating an answer. A teammate you can’t trust to say “I’m not sure” isn’t useful.

What it does with that knowledge

Section titled “What it does with that knowledge”
  • Answers technical questions with your team’s real context: systems, customers, past incidents.
  • Records repeatable fixes as playbooks so the next person doesn’t start from scratch (see Playbooks).
  • Asks the right owner before running anything destructive (see Human approvals).
  • Nudges people to close out incidents and tasks that are slipping.

It can also automate routine work like scheduled summaries and recurring checks, but the value is having a knowledgeable teammate in the room, not another automation to configure.

Next steps

Section titled “Next steps”