Thinkster + Noteriv🔗

Thoughts and Tangents🔗

Most thinking isn’t linear. It starts with an observation, splits into a tangent, splinters into a sub-tangent, and eventually needs to be consolidated and merged back into the main line of thought—much like a git branch.

I conceived Thinkster as an editor designed specifically around this recursive branching workflow: take a thought, spin it into a branch, explore the rabbit hole, and merge the synthesized conclusions back when you hit a resting point.

Things I’m Testing🔗

1. Dense Local Models for High-Velocity Ideation🔗

One immediate realization: for 80% of iterative drafting, outlining, and bouncing ideas, you don’t need a frontier reasoning model or a trillion-parameter cloud endpoint with network latency and per-token meter anxiety.

• Most testing runs against dense, quantized local models fitting comfortably under 20GB of VRAM.
• Running locally at ~66 tokens/second delivers instantaneous feedback loops that keep cognitive momentum unbroken.

2. The Limits of Terminal Real Estate🔗

Thinkster started purely as a CLI markdown tool, optimized for fast keyboard navigation and zero context switching. But terminal real estate quickly became a hard bottleneck:

• Handling deep recursive tangents and tool traces side-by-side collapses on standard 80–120 column terminals.
• Markdown prose and nested blockquotes consume significant vertical height, causing previous context to scroll out of view.

3.Spatial UI with Noteriv & Andy Matuschak Panes🔗

To solve the spatial layout problem without abandoning keyboard-centric agility:

• Explored porting the branching engine into Noteriv, a fast open-source Tauri-based note taker built in the spirit of Obsidian.
• Implemented Andy Matuschak-style cascading sliding panes, allowing tangents to open horizontally across layered panes. You can trace your way five levels deep into a research tangent, pan smoothly across your train of thought, and collapse back to root without losing spatial awareness.