Proposal to enhance the webhook API

[gimelg]

Greetings!

I'd like to suggest an update to the Webhook API (demo below) that would let us build more familiar, interactive chat experiences like those seen in modern LLM interfaces. The goal is to enhance the user experience by introducing richer message types that convey the model's activity and output more naturally.

Thinking Messages

A new type of ephemeral message intended to visually communicate that the assistant is processing. These messages are not persisted to the database, but serve as real-time feedback to the user.

• Ephemeral, used only to reflect ongoing computation.
• Can be updated continuously as progress evolves (e.g. "Analyzing…", "Fetching context…", "Generating result…").
• Helps set expectations during longer-running tasks.

Animated Messages

Messages that behave normally, but are delivered in a live typing / streaming style, similar to the UX people are accustomed to with ChatGPT and other LLM interfaces.

• Persisted like normal bot messages.
• Appear to be streamed to the UI in real-time to simulate typing or gradual output.
• Ideal for long responses or when progressive rendering improves UX.

Screen recording demonstrating usage:

chatbot_demo.mp4

Implementation

I've built the following spec off of which I implemented the feature:
BOT_COMMUNICATION_ENHANCEMENTS_SPEC.md

Happy to iterate further or send a PR if you agree this is useful to have.

Cheers,
G