Most AI agents are black boxes: prompt in, result out.
For the #meliuschallenge, I built Nuncio — an agentic video personalization pipeline where the agent’s reasoning unfolds live on an interactive @Melius canvas.
How it works:
1️⃣ Context → TinyFish pulls recipient social profiles in parallel
2️⃣ Synthesis → Claude selects a “visual hook archetype” (Mirror, Origin, Future-cast) and scripts a personalized pitch
3️⃣ Creation → The agent uses Melius’s MCP server to programmatically build a canvas layout, wiring nodes and edges into visual data flows
4️⃣ Rendering → Output is stitched with a HeyGen Avatar V talking head and Speechmatics-timed captions
The Melius canvas isn’t just where assets live — it’s the agent’s canvas of thought. Users can then use that brainstorm pattern to generate highly customised outreach videos.
https://nuncio.persidian.com/
https://github.com/udirobert/nuncio
0 likes · 0 views · May 20, 2026
Most AI agents are black boxes: prompt in, result out. For the #meliuschallenge, I built Nuncio — an agentic video personalization pipeline where the agent’s...