Roadmap

[kommander]

Build the features OpenCode needs, stabilize the core, then refactor and optimize.

Now

v0.1, v0.2, v0.3 — Exploring the problem space.

• [DONE] CLI render mode (split mode), with interactive TUI at the bottom and immutable scrollback log stream at the top.bf8f195
• [DONE] Rework color transport for RGBA+A and indexed metadata. Cache nearest-index translations so both TS and native paths skip repeated conversion. 95d36f3
• [DONE] Split the 3d/threejs core part into a separate package a10e71c
• Find what features OpenCode and terminal.shop need
• Replace setStyledText with a native interface for incremental text updates (depends on MarkTree)
• Cross-platform clipboard with OSC52 and platform-specific fallbacks
• Make OpenTUI run on other runtimes like Node and Deno (debunification)
• Native streaming path for SSH connections
• Add virtual text for textarea autocompletion. TypeScript-only first, native via MarkTree later
• Hunk navigation for code review in OpenCode. Inline editing likely depends on MarkTree
• Timing-based scroll handling with inertia and speed, telling trackpad from mouse. Partial version exists
• Load external modules at runtime (in progress)
• OpenCode tests use OpenTUI test helpers. Fill gaps in OpenTUI's testing tools
• Kitty Graphics & Sixel support

Next

v0.x — Refactor the hell out of it.

• Neovim-style sticky marks (MarkTree/extmarks) for highlights, virtual text, and conceal. Marks move with edits so treesitter skips full re-highlighting
• Move markdown parsing and rendering to native code, compatible with MarkTree
• TextBufferView caches all virtual lines, even invisible ones. Constrain to the viewport
• Close the gap between OpenTUI's grapheme/width model, Unicode, and how terminals render complex text
• Kitty keyboard protocol follow-ups for repeat event shape, modified Tab encoding, KP_BEGIN naming, and other spec gaps

Later

v1.0 — Everything moves native.

• The renderable tree, layout engine (Yoga), and rendering move to native code. Native callbacks for calculateLayout let you update renderables directly without a separate update pass

Ongoing

Work that spans every phase.

• Replace inconsistent TS benchmarks with a canonical setup. Go-style calibration, warmup, robust stats, and JSONL output for opentui-bench
• Yoga-via-wasm has large overhead. Build Yoga into the OpenTUI binary with manual bindings. Becomes obsolete when the renderable tree goes native
• Rope is overkill for read-only text. A simpler TextBuffer would reduce overhead. Deprioritized after mixed benchmarks (Feb 2026)
• Remove non-determinism from the frame pipeline with an injectable Clock abstraction (in progress)

[viralcodex]

One question on the roadmap @simonklee, will most of the code in the coming version will slowly move from TS to zig? curious coz of 1.0 means everything moved to native.

[simonklee]

There are some things that benefit more from being in native than others. Then there are things where you need some sort of event loop for it work. Finally, you'll probably always have parts that are only in ts.

Current focus is not to move things to native, though.

[joshmeads]

Then there are things where you need some sort of event loop for it work.

While it's not in the current version of Zig, Andrew did share his vision for the new async work he's doing which includes std.Io.EventLoop.

https://gist.github.com/andrewrk/1ad9d705ce6046fca76b4cb1220b3c53

Unsure of its progress, he did mention a new async will be one of the larger breaking changes to Zig - so I doubt they're rushing it.

Ultimately Bun is fast and has its own native event loop so I agree there isn't a whole lot of benefit in OpenTuis case of moving it at the moment.

[jettptacek]

Do you guys want outside help working on these items?