project file / jun–jul 2026
Tangent
A workspace for branching an AI conversation, keeping the source context of each tangent visible, and returning with an insight.
- my role
- Independent builder
- status
- Local-first prototype
- field
- Personal prototype / interfaces
- tools
- JavaScript · CSS · Node.js · Local persistence
01 / the problem
Why this exists
Linear chat is awkward when one sentence opens several useful questions. Tangent explores how to follow a branch without losing the original conversation or the reason the branch exists.
02 / the build
What I worked on
Built a responsive conversation workspace with selectable tangents, a navigable graph, editing, tags, search, and local persistence.
Added branch context packets, version snapshots, JSON import and export, and a small Node backend for local storage and simulated streaming runs.
Designed mobile navigation and recovery paths for deleting or restoring branches.
03 / engineering notes
Choices that mattered
- A branch keeps its origin and context instead of becoming an isolated new chat.
- The committed prototype uses a mock model provider. It demonstrates the interaction model without claiming production authentication or live-model reliability.
04 / reflection
What I took away
Tangent is where I could be more exploratory about interface design. The useful result so far is a clearer model for how a conversation can branch and still remain understandable.