01 shipped
AI Customer Support
The AI backline for customer support.
Killer signal
See it live
“This bug has been reported five times across five tickets this month.”
Use cases · Gallery
Same loop. Different channel. Infinite use cases.
The pattern is simple: listen to a stream, classify, research, draft for a human, build memory. Every card below is that loop with a different input. Support was first. These are next.
The meta-pattern
One loop, many jobs.
Daimon is really a template for building an AI coworker that watches a stream, drafts for a human, and builds memory over time. Customer support is one instance. PR review, on-call, community, sales ops, product analytics — all the same shape, different channel, different routes.
The gallery
Twelve places Daimon fits.
02 reference fork planned
PR Review daimon
The AI staff engineer who never forgets why you did it that way.
Killer signal
“Three PRs this month touched the same middleware. Two had the same bug.”
03 idea
On-call SRE daimon
The AI incident historian.
Killer signal
“This alert pattern is the 3rd of its kind in 90 days.”
04 idea
Community daimon
A mod that's been here since day one.
Killer signal
“Six people asked the same thing in 10 days and it's not in the docs.”
05 idea
Product analytics daimon
A daily read of what users actually did.
Killer signal
“No one used the new export this week, and sessions on the page it lives in didn't drop.”
06 idea
Sales ops daimon
A sales assistant that actually follows through.
Killer signal
“This prospect asked about X integration three times — same question we've filed twice before.”
07 idea
Investor relations daimon
The update-writer who never forgets what you promised.
Killer signal
“You told the board you'd hit X by end of Q2. Week 10. Hasn't come up in standup once.”
08 idea
Compliance daimon
The thing that notices before legal does.
Killer signal
“Someone added an SDK that reads user data. No privacy review on the PR.”
09 shipped · the original skeleton
Personal journaling daimon
A persistent AI that lives in your family telegram group.
Killer signal
“The demo's come up four times this week — three mentions were 'avoiding it'.”
10 idea
Content / SEO daimon
A writer-in-residence that reads your analytics.
Killer signal
“Three top-10 posts haven't been updated in 18 months and rankings are slipping.”
11 idea
Recruiting daimon
Pipeline memory that survives the recruiter leaving.
Killer signal
“This candidate applied 8 months ago and got a strong 'not now.' Worth revisiting?”
12 idea
Legal / contract daimon
Reads renewal dates so you don't get auto-renewed at 2×.
Killer signal
“Your vendor updated their DPA last month. Here's the diff. One clause now says X.”
The shape
Where Daimon fits.
Daimon is strongest where four things line up: a stream of inputs over time, answers that need cross-referencing and memory, value in pattern detection over instant reply, and a human who wants draft + signal, not autonomous action.
- ✓ Stream of inputs over time
- ✓ Answers that need cross-referencing
- ✓ Patterns matter more than instant reply
- ✓ Human-in-the-loop is a feature, not a bug
If three or four of these describe your problem, Daimon probably fits. Open an issue on GitHub and tell us about it.
Build your own
The skeleton ships with everything you need.
Engine
- Persistent systemd session
- Git-tracked markdown memory
- Dream-pass scheduler
- Noticing function (absence, trend, connection)
- Test harness
I/O
- Telegram plugin (reference)
- Channel abstraction — swap for Intercom, Discord, Slack, GitHub
- MCP server support
Ops
bin/install— one-command installbin/upgrade— tagged upgradesbin/doctor— health check- Template + render system for systemd units
Build one. Share it back.
Built a daimon for something new? Open a PR on the skeleton with your module as a reference fork. We'd love to link to it from here.