mirror of
https://github.com/supermemoryai/supermemory.git
synced 2026-08-28 05:25:33 +00:00
97 lines
4.9 KiB
Text
97 lines
4.9 KiB
Text
---
|
|
title: "Automations and Proactiveness"
|
|
sidebarTitle: "Automations"
|
|
description: "Scheduled work Company Brain runs on its own, and when it speaks without being asked"
|
|
icon: "bot"
|
|
---
|
|
|
|
import { SlackThread, SlackMessage, Mention, ChannelRef } from "/snippets/slack-message.mdx";
|
|
|
|
Company Brain doesn't only answer when you @mention it. It can run recurring work on a schedule, and it can speak in a thread on its own when it has something genuinely worth saying. Both are opt-in, both are rate-limited, and both read from exactly the same [permissions graph](/company-brain/permissions) as a normal question — neither is a backdoor around it.
|
|
|
|
## Automations
|
|
|
|
An automation is a prompt that runs on a schedule and posts the result somewhere. You write it once, in plain language:
|
|
|
|
<SlackThread channel="#product">
|
|
<SlackMessage self time="9:03 AM">
|
|
<Mention>supermemory</Mention> every Monday at 9am, post a digest of what shipped last week and what's still open, to <ChannelRef>product</ChannelRef>.
|
|
</SlackMessage>
|
|
<SlackMessage bot time="9:03 AM">
|
|
Got it — scheduled. First digest posts Monday, 9:00 AM, to <ChannelRef>product</ChannelRef>.
|
|
</SlackMessage>
|
|
</SlackThread>
|
|
|
|
<Steps>
|
|
<Step title="Schedule fires">
|
|
The automation wakes up at its set time — no one has to trigger it.
|
|
</Step>
|
|
<Step title="Gathers context">
|
|
It reads using only **org-shared** connections and channel memory — never a person's personal credentials, even if the person who created the automation has better personal access. This is what keeps a scheduled post from silently acting as a specific teammate.
|
|
</Step>
|
|
<Step title="Checks visibility">
|
|
Before posting, it re-confirms it can still see the destination channel.
|
|
</Step>
|
|
<Step title="Posts, or fails closed">
|
|
If anything above is unclear — a connection broke, visibility can't be verified — it skips that run rather than posting a guess. Silence beats a wrong digest.
|
|
</Step>
|
|
</Steps>
|
|
|
|
**Who can target what:**
|
|
|
|
| Destination | Who can create it | Reads from |
|
|
|---|---|---|
|
|
| Public channel | Any member | Org-shared connections, public channel memory |
|
|
| Private channel | Admins only | Org-shared connections, that channel's memory |
|
|
| DM to yourself | The owner of that DM | Your personal + org connections, your employee memory |
|
|
|
|
Common shapes worth stealing:
|
|
|
|
- A Monday-morning digest of open items and unanswered questions
|
|
- A daily Sentry error recap in `#eng`
|
|
- A weekly "what changed across our connected tools" summary
|
|
|
|
Anyone can create and manage their own automations; admins can manage everyone's. Ask Company Brain in Slack to set one up, or manage the full list from the web app.
|
|
|
|
## Proactiveness (chime-in)
|
|
|
|
Chime-in is different from an automation: there's no schedule, and no one asked. Company Brain is simply present in a channel — because an admin invited it — and it speaks up when staying quiet would waste someone's time.
|
|
|
|
**What actually earns a chime-in:**
|
|
|
|
- It has to add something the room doesn't already have — a fact, a correction, a next step — not agreement or a restatement of what's already visible.
|
|
- It has to come from somewhere it's genuinely allowed to look: [connected tools](/company-brain/connectors) or that room's own memory, same as any other answer.
|
|
- If it isn't confident the answer is actually correct, it says nothing. A wrong guess is worse than silence, so uncertainty resolves to silence, not a hedge.
|
|
|
|
<CodeGroup>
|
|
```text Worth chiming in
|
|
"is prod down? customers are pinging me"
|
|
→ correlates against Sentry, replies with what's actually elevated right now
|
|
```
|
|
|
|
```text Not worth it
|
|
"finally shipped this 🎉" (screenshot, no question)
|
|
→ stays quiet — there's nothing to add
|
|
```
|
|
</CodeGroup>
|
|
|
|
**Guardrails that keep it from becoming noise:**
|
|
|
|
- **Rate-limited.** It won't speak repeatedly in the same thread or channel in a short window, even if it technically could add something each time.
|
|
- **Invite-only rooms.** It never joins a channel on its own — only places an admin already invited it into.
|
|
- **Same graph as a normal answer.** A private channel's chime-in only ever draws on that channel's memory and public channel memory — never another private channel, never someone else's employee memory.
|
|
|
|
An explicit @mention always skips this judgment call entirely — naming it is you deciding it should speak, so it does.
|
|
|
|
<Note>
|
|
Automations and chime-in both write back to memory the same way a normal conversation does: a public channel's automation output lands in public channel memory, a private channel's chime-in stays scoped to that channel's memory.
|
|
</Note>
|
|
|
|
<CardGroup cols={2}>
|
|
<Card title="What you can do" icon="sparkles" href="/company-brain/use-cases/overview">
|
|
Real scenarios — support, incidents, digests, and more.
|
|
</Card>
|
|
<Card title="Connectors" icon="plug" href="/company-brain/connectors">
|
|
Wire up the tools automations and chime-in draw from.
|
|
</Card>
|
|
</CardGroup>
|