Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.openori.dev/llms.txt

Use this file to discover all available pages before exploring further.

OpenOri isn’t just reactive. It runs a background intelligence cycle that monitors your connected services, fires reminders on time, and learns what’s worth your attention — all silently, without interrupting your flow.

​
How it works

The ambient system has three layers: 
┌────────────────────────────────────────────┐
│           Your connected services          │
│    (GitHub, Gmail, Calendar, Slack, ...)    │
└──────────────────┬─────────────────────────┘
                   │
         ┌─────────▼─────────┐
         │   Ambient Cycle   │  ← runs every 5 min
         │   (same agent)    │
         └─────────┬─────────┘
                   │
    ┌──────────────┼──────────────┐
    ▼              ▼              ▼
 Intents      Connector       Learning
 (reminders)  observation     (triage)
  1. Intent firing — checks every 5 seconds for due reminders and delivers them instantly
  2. Connector observation — every 5 minutes, scans your connected services for things worth surfacing
  3. Triage learning — tracks what you engage with vs. dismiss to improve relevance over time

​
The ambient cycle

The ambient cycle uses the same agent pipeline as your regular conversations — same model, same tools, same Skill Graph routing. The only difference is the prompt, which includes:
  • Current time and time zone
  • Due reminders and pending intents
  • Connected services to check
  • Learned triage hints (what you care about)
The agent decides what’s worth surfacing and produces ambient work items — lightweight notifications that appear as toast messages.

​
Ambient work items

When the cycle finds something noteworthy, it creates an ambient work item with one of three intents:
IntentWhen usedExample
InformFYI, no action needed”3 new stars on open-ori/openori today”
RequestNeeds your attention”Reminder: Walk the dog”
ReportDetailed result for review”Weekly PR digest: 12 merged, 3 open”
Each item can include action buttons like Got it, Snooze, or Dismiss.

​
Smart timing

The cycle adapts to your schedule:
TimeBehavior
Normal hours (7AM–11PM)Full cycle every 5 minutes
Quiet hours (11PM–7AM)Skips cycles unless there are due reminders
After idle (10+ min away)Burst-on-focus: runs immediately when you return
Burst-on-focus means when you come back to OpenOri after a break, you see fresh notifications immediately — no waiting for the next 5-minute tick.

​
Triage learning

The ambient system learns from your behavior:
  • Engaged — you clicked, replied, or acted on a notification → more of these
  • Dismissed — you swiped it away → fewer of these
  • Snoozed — you wanted it later → adjust timing
Over time, OpenOri gets better at knowing what’s worth interrupting you for and what can wait.

​
Using it

You don’t configure the ambient system directly — it runs automatically once you’ve connected services or set up reminders. Here’s what activates it:
You do this…Ambient system does this…
Connect GitHubMonitors PRs, issues, and notifications
Connect GmailWatches for important emails
Connect CalendarAlerts you before meetings
Set a reminder (“remind me in 30 min”)Fires a toast + OS notification at the right time
Create a scheduled taskRuns it on schedule in the background

​
Privacy

The ambient cycle uses your chosen AI provider — the same model you chat with. If you use Ollama, the entire ambient system runs offline on your hardware. No data is stored externally. Ambient work items are kept in ~/.ori/ambient-work.json and auto-expire after 24 hours.
The ambient system is completely transparent. Check the terminal output for [ambient-cycle] logs to see exactly what it’s doing.