Image input via Telegram
A photo of the striping machine's spray tip was sent to tito over Telegram for classification.
OpenClaw identified it as a 515 tip, good for wide coverage but wrong for 4-inch lines. The correct tip is a 319. Telegram forwards the image as a file, OpenClaw includes it in context, and the model can reason over it without extra setup.
Research and CRM write
The next task was to research airless sprayer tip sizing and write the result to the CRM:
- Search for spray tip charts and specs.
- Create a structured guide covering orifice size, fan width, pressure, and uses.
- Post to the CRM as a note:
bashcurl -X POST https://striping-app.<account>.workers.dev/api/contacts/<id>/notes \
-H "Authorization: Bearer $CRM_API_KEY" \
-H "Content-Type: application/json" \
-d '{"body": "Tip sizing guide: 315 for 4" lines, 519 for broad coverage..."}'
The note was written to a contact's activity feed. Day 2 still used direct curl calls. Python tooling came later.
Validated on day 2
Three things worked without any new setup:
- Multimodal input: Telegram images integrate directly.
- Web research: The agent synthesizes external content.
- CRM write: Day 1's credentials authenticated note writes.