July was a month about removing steps. Nearly every item below started life as a support ticket about Tagni itself, which is a slightly humbling way to build a support product, and a very effective one.
Onboarding, rebuilt around the first answer
The old setup flow asked you to name your workspace, invite your team, pick a plan and connect a channel before Tagni had said a single useful thing. People were dropping out at step three, and fairly.
The new flow asks for one thing: a URL. Tagni crawls it, builds a draft knowledge engine, and puts a working agent in front of you inside about ninety seconds. Everything else, team, billing, channels, moved to after the moment you can see it working.
Median time from signup to first correct answer went from just over eleven minutes to under two.
Files that actually get read
Agents can now read PDFs, spreadsheets and images wherever they arrive, chat, email or WhatsApp. That matters more than it sounds. A shipping dispute usually arrives as a photo of a damaged box; a billing question arrives as a PDF invoice someone exported months ago.
- PDFs are parsed with layout preserved, so tables survive the trip
- Spreadsheets are read per-sheet, with headers detected
- Images are described and, where there's text, transcribed
If a customer attaches an invoice and asks "why is this different from last month", the agent can now answer without a human opening the file first.
A one-click site plugin
Adding the chat widget used to mean pasting a script tag, which meant finding someone who could deploy. The new plugin installs from your site builder directly, pick the workspace, choose a position, publish.
The whole point of a support tool is that it removes work. Setup is work.
Smaller things worth mentioning
- Draft replies now respect per-agent tone settings instead of a single global voice
- Reporting exports as CSV, finally
- Threads can be merged when the same customer emails twice
- Webhook retries use exponential backoff, which fixed the duplicate-event reports several of you sent in
What's next
August is about escalation. Right now handing a conversation from an agent to a person is clean, but handing it back isn't. We want an agent to be able to pick a thread back up once the human has unblocked it, with the full context of what changed in between.