Cursor Automations
Scheduled cloud agents for daily maintenance and edition server management
Overview
Cursor Automations (available since March 5, 2026) run cloud agents on a schedule — no local launchd configuration, no XML plists, no cron jobs on your Mac.
Why automations complement launchd:
- Run even when the Mac is sleeping or offline
- Agent memory: learns from past failures, skips already-processed items on retry
- Prompt-driven: adjust behaviour by editing text, not XML
- Health-check capable: each run can verify the previous state before acting
- Unified audit trail at cursor.com/automations
Create or edit automations at cursor.com/automations/new.
INQUA services are off-limits. Never reference or restart ports 8080, 5101, or any service under ~/inqua2026/ in any automation.
Edition Servers Inventory
All scholarly edition servers are ai-viewer instances. Start commands live in config/service-commands.yaml. None have launchd plists — they are started by the daily restart automation below.
| Route | Port | Edition | Domain | Start command |
|---|---|---|---|---|
/euler-opera |
5103 | Euler Opera Omnia | euler-opera.lettre.ai | ai-viewer --server --port 5103 |
/kepler-gw |
5104 | Kepler Gesammelte Werke | kepler-gw.lettre.ai | ai-viewer --server --port 5104 |
/copernicus |
5105 | Copernicus Studies | copernicus.lettre.ai | ai-viewer --server --port 5105 |
/waefler |
5106 | Waefler-Hamidi Archive | waefler.lettre.ai | ai-viewer --server --port 5106 |
/kirch |
5109 | Kirch Correspondence | kirch.lettre.ai | ai-viewer --server --port 5109 |
/sundials |
5720 | Ancient Sundials (BSDP) | ancient-sundials.lettre.ai | ai-viewer --server --port 5720 |
/karman |
5725 | Von Kármán Archive | karman.lettre.ai | no web server (CLI tool only) |
/ancient-sundial |
2200 | Sundial Shadow Animator | ancient-sundial.cooltool.io | uvicorn sundial_server:app --port 2200 |
The enestroem collection is served through the general /viewer instance (port 5100).
Executable: /opt/homebrew/bin/ai-viewer — no conda env needed
Working directory: ~/Dropbox/202x/2026project/ai-viewer (except ancient-sundial which uses ~/Dropbox/202x/2026research/ai-sundial3d)
Automation 1: Edition Servers Daily Restart
Restarts all scholarly edition servers each morning before work hours.
| Setting | Value |
|---|---|
| URL | cursor.com/automations/new |
| Trigger | Scheduled — 0 5 * * * (05:00 daily) |
| Repository | ai-platform |
| Tools | Memories |
| Environment | Disabled (no dependency install needed) |
Prompt:
You are the Edition Servers restart agent. Each morning, restart the
following scholarly edition servers. Start commands are in:
~/Dropbox/202x/2026project/ai-platform/config/service-commands.yaml
For each edition below:
a) Kill any existing process on its port:
lsof -ti :PORT | xargs kill -9 2>/dev/null || true
b) Start using conda run with the voicememo-py314 environment
c) Wait 5 seconds, then verify: curl -sf http://localhost:PORT/ > /dev/null
d) Log success or failure in memory
Editions to restart (in order):
- euler-opera port 5103 cwd: ~/Dropbox/202x/2026project/ai-viewer
cmd: /opt/homebrew/bin/ai-viewer --server --port 5103
- kepler-gw port 5104 cwd: ~/Dropbox/202x/2026project/ai-viewer
cmd: /opt/homebrew/bin/ai-viewer --server --port 5104
- sundials port 5720 cwd: ~/Dropbox/202x/2026project/ai-viewer
cmd: /opt/homebrew/bin/ai-viewer --server --port 5720
- ancient-sundial port 2200 cwd: ~/Dropbox/202x/2026research/ai-sundial3d
cmd: /opt/homebrew/bin/python3 -m uvicorn sundial_server:app
--host 0.0.0.0 --port 2200
Note: karman (port 5725) has no web server — skip it.
CRITICAL: Do NOT touch ports 8080, 5101, 5200, 5001 — those are INQUA and core services.
After all restarts, write a status report to:
~/Dropbox/202x/2026project/ai-platform/logs/editions_restart_YYYY-MM-DD.txtAutomation 2: PrivatKanzlei Daily Update
Runs all three PrivatKanzlei data sync jobs in coordinated sequence.
| Setting | Value |
|---|---|
| URL | cursor.com/automations/new |
| Trigger | Scheduled — 0 6 * * * (06:00 daily, after nightly launchd jobs finish) |
| Repository | PrivatKanzlei (or ai-platform) |
| Tools | Memories, MCP user-korrespondenz |
| Environment | Disabled |
Prompt:
You are the PrivatKanzlei daily update agent. Each morning, run all three
update jobs in sequence in ~/Dropbox/ggnew/PrivatKanzlei:
1. Gmail refresh:
python3 scripts/gmail_auto_refresh.py --once
Confirm OAuth tokens are valid.
2. Nightly sync:
python3.11 scripts/nightly_mail_calendar_sync.py
Syncs Gmail + Thunderbird mail, extracts calendar events via Gemini,
pushes to Google Calendar.
3. Belege auto:
python3 packages/korrespondenz/tools/firma_belege_auto.py
Processes new company receipts/documents.
After each step check the exit code. If a step fails, record the error in
memory and continue. At the end write a summary (mails synced, calendar
events created, Belege processed) to:
~/.ai-korrespondenz/logs/daily_summary_YYYY-MM-DD.txtAutomation 3: VoiceMemo Daily Pipeline
Runs the full VoiceMemo pipeline: import → transcribe → classify → report update.
| Setting | Value |
|---|---|
| URL | cursor.com/automations/new |
| Trigger | Scheduled — 0 7 * * * (07:00 daily) |
| Repository | ai-voicememo |
| Tools | Memories |
| Environment | Enabled (needs Python deps) |
Prompt:
You are the VoiceMemo daily pipeline agent. Run the full pipeline each
morning from ~/Dropbox/202x/2026project/ai-voicememo:
1. Import:
python scripts/import_recent_memos.py
Scan Apple Voice Memos for new recordings, import to local DB.
2. Transcribe:
python scripts/sync_and_process.py
Transcribe pending memos using Gemini.
3. Classify:
python scripts/classify_memos.py
Classify transcripts into the correct project.
4. Report index:
python scripts/generate_report_index.py
quarto render docs/reports/
Update the Quarto report site.
5. Validate:
python scripts/validate_all_transcripts.py
Report missing or failed transcripts in memory.
On failure, use memory to skip already-processed memos on the next run.
Write a summary to:
~/Dropbox/202x/2026project/ai-voicememo/logs/daily_summary_YYYY-MM-DD.txtSetup Checklist
Before the automations run for the first time:
- GEMINI_API_KEY secret — add at cursor.com/dashboard → Cloud Agents → Secrets
- Onboard repositories — each automation requires a repository to be onboarded at cursor.com/onboard
- Memory initialisation — on first run, add a memory note: “First run — no prior state”
- Verify cron expressions — double-check scheduling at crontab.guru
Daily Schedule Overview
| Time | What runs | Type |
|---|---|---|
| 02:00 | INQUA report update | launchd |
| 02:30 | PrivatKanzlei nightly sync | launchd |
| 03:00 | Korrespondenz sync | launchd |
| 04:00 | Firma Belege auto | launchd |
| 05:00 | Edition Servers Restart | Cursor Automation |
| 06:00 | PrivatKanzlei Daily Update | Cursor Automation |
| 07:00 | VoiceMemo Daily Pipeline | Cursor Automation |
| 08:00–21:00 | VoiceMemo hourly sync | launchd |
Monitoring
View run history, logs, and memory state at cursor.com/automations.
Each automation writes a local summary log:
| Automation | Log path |
|---|---|
| Edition Servers Restart | ~/Dropbox/202x/2026project/ai-platform/logs/editions_restart_YYYY-MM-DD.txt |
| PrivatKanzlei Daily Update | ~/.ai-korrespondenz/logs/daily_summary_YYYY-MM-DD.txt |
| VoiceMemo Daily Pipeline | ~/Dropbox/202x/2026project/ai-voicememo/logs/daily_summary_YYYY-MM-DD.txt |
Retiring Redundant launchd Plists (Optional)
Once the Cursor Automations have run stably for two weeks, consider retiring these two PrivatKanzlei launchd plists to avoid double-running:
launchctl unload ~/Library/LaunchAgents/com.privatkanzlei.nightly-sync.plist
launchctl unload ~/Library/LaunchAgents/com.opensciencetechnology.firma-belege-auto.plistKeep all other launchd plists (especially all INQUA services) unchanged.