Getting started
Auto-summons: the same agent, no tag required
Point your monitoring at Novix and an error your users hit becomes a diagnosed task and a pull request before anybody reports it.
A person summons Novix by writing in or naming it on a thread. An auto-summon starts from a product signal instead. Nothing after that changes: it is the same Novix agent, the same nine stages, and the same human approval boundary.
What an auto-summon joins
Three things, and the loop only closes with all three. The first two you already have; the third is the one this doorway alone needs, because an alert arrives with no thread, no reporter and nobody waiting for a reply.
- A source. PostHog, Sentry, Datadog, Rollbar, PagerDuty or Statuspage. PostHog reads the exceptions your users hit, and nothing else: there is no path from what Novix asks it for to a pageview, a funnel or anybody’s session.
- A codebase. Which repository the alert is diagnosed against. Leave it automatic and Novix routes each one the way it routes every other task.
- A destination. Where the answer is published, because there is no thread to answer in. There is no default: guessing would put a diagnosis somewhere nobody picked.
Where the answer can go
| Destination | What Novix creates there |
|---|---|
| Slack | A thread. Answer in it and your reply comes back to the task. |
| Discord | A message. |
| Linear | An issue. |
| Notion | A page. |
| A message. Nothing to connect, so it works on day one. |
Slack is the only two-way one. Everywhere else the answer is published and any reply belongs on the task.
The three moments it speaks
- The answer. What Novix found, and the pull request if it drafted one.
- A question. When it needs something before it can go further.
- A failure. When the run could not finish. Finding no fix is an answer, not a failure.
A question and a failure go wherever the answer goes unless you send them somewhere else.
Only defects become tasks
Nobody chose to file any of this, so a stream that opened a task per recovery, acknowledgement and test page would fill your inbox with rows no human will ever close. Novix screens twice: once before the run, on the alert itself, and again afterwards on its own verdict, deleting the task if it was not a defect. A task waiting on an answer from you is always kept.
What your plan includes
Solo runs one auto-summon, start to finish: monitoring in, the same Novix agent diagnoses against your code, a pull request comes out, and the answer is published to one place you chose. Nothing about the loop is degraded. Pro describes several and shapes each one: which alerts are worth a deep run, sparing a second run when the same alert repeats, a separate place for a question, a daily cap on runs, and a quiet window. See the plans.
A workspace that moves down a plan keeps hearing about its outages on the routing it already set up. Only the next save is clamped.