Use one interaction across tools
Provide one fuzzy-search picker and familiar direct launchers for each supported provider.
↳ Documented product behavior← Selected work
AI workflows · Developer experience / 2026
Agent Session Hub
A shared session picker for Codex CLI, Claude Code and OpenCode, with previews, aliases and project-aware resume.
My role
System & interaction design / development
The scope
Rust · CLI design · AI workflows
Project status
Released cross-platform CLI
The project, at a glance
- The challenge
- AI coding sessions become fragmented across providers and folders, making the right context difficult to recover.
- My contribution
- I designed and built a local search and preview workflow, with Rust session parsing, terminal interaction and shell integration.
- What came out of it
- Public release artifacts for macOS, Linux and Windows, connecting session discovery to continued work.
Developer experience / Agent Session HubFind the conversation.
Find the conversation.
Restore the context.
01Search across tools02Inspect the project03Resume in context
New workflow illustration · Fictional session labels
⌕ Search sessions ⌘ K
THREE TOOLS. ONE PLACE TO CONTINUE.
CodexPortfolio interface./portfolio
Claude CodeComponent library./design-system
OpenCodeApp integration./workspace
PREVIEW → PROJECT → RESUME
Pick up the thread.
Keep the context.
Workflow illustration / fictional sessions
Agent Session HubExisting project artwork
01 / The question behind the work
Pick up the right AI conversation, in the right project.
The problem
AI coding work becomes fragmented across providers, project folders and long session histories. Finding the right conversation is only half the job; it also needs to reopen in the right context.
My contribution
Cross-tool workflow design, Rust session parsing, terminal interaction, search/preview experience and shell integration.
What the work delivers
Native Rust CLI, Codex / Claude Code / OpenCode parsers, fzf browser and preview, Alias and session management, Shell integration, macOS, Linux and Windows release artifacts.
02 / Design decisionsThe thinking
The thinking
in the details.
Specific choices, connected to the work you can see.
Keep the project attached
Resume sessions from the matching working directory where possible; expose repository and branch context in discovery.
↳ Documented product behaviorMake cleanup deliberate
Group generated agents below conversations, preview storage implications and require explicit confirmation for deletion.
↳ Documented product behavior03 / Connecting the experience
How the pieces connect.
Select a stage to explore its purpose.
The moment
Search conversations across supported tools.
The design intention
Provide one consistent entry point.
New portfolio explanation based on the documented implementation and original project material.
04 / Inside the experienceThe details.
The details.
With a purpose.
Selected views, framed to explain the work. Complete original artifacts are available in the source collection below.
⌕ Search sessions ⌘ K
THREE TOOLS. ONE PLACE TO CONTINUE.
CodexPortfolio interface./portfolio
Claude CodeComponent library./design-system
OpenCodeApp integration./workspace
PREVIEW → PROJECT → RESUME
Pick up the thread.
Keep the context.
Workflow illustration / fictional sessions
One workflow across tools
This new illustration explains search, preview and project-aware resume. The labels are fictional; no private session history is presented.
Original presentation boards1 source artifacts +
The complete originals behind this case study. Open a board to inspect the source presentation.
05 / What this work brings togetherAgent Session Hub
Agent Session Hub
in perspective.
The result
Public v0.3.10 release artifacts were verified for macOS, Linux and Windows. No adoption or time-saving metric was established.
Sources & presentation notes +
Existing banner is artwork. Use a labelled workflow illustration if adding a new terminal presentation. This research did not open or screenshot the owner’s real AI session history.
The journey diagrams and editorial explanations were prepared for this portfolio from the available project evidence. They are not additional historical research or measured outcomes.
Keep exploring / Next project
An experimental memory layer that combines a local proxy, encrypted storage and a bounded execution model with traceable assertions.