Browser-based graphical shell and control plane for a Linux server you own — real terminal, file manager, live metrics and BYOK AI in one mobile-first workspace. Next.js 16 · React 19 · node-pty · Tailwind v4 · shadcn/ui. Public alpha, self-hosted.
TUI parity plan (opencode → mso)
MSO Agent TUI parity program contract. The Status column in the tables below is the single source of truth for what shipped. Every item names the seam it uses. An item with no real seam is marked NOT DONE instead of being faked.
Audit baseline
Measured from source, not from screenshots.
- mso: 30 modules, 3687 lines in
scripts/mso-agent-*.mjs. Plain ANSI writes. No alt-screen (\x1b[?1049), no mouse tracking, no panel/scrollbar engine. - opencode: Bun binary, OpenTUI + Solid renderer.
scrollbox,scrollbarOptions, mouse handlers, session tabs, diff viewer, command palette, sounds, OS notifications, configurable keybinds.
The renderer gap is the root of most visual gaps. Everything below avoids it unless a phase explicitly pays for it.
Phase 0: cheap wins on existing seams
| # | Item | Seam | Status |
|---|---|---|---|
| P0-1 | /export [ref] [--out f] in the TUI |
mso session export (already shipped) |
shipped |
| P0-2 | /share authenticated temporary link for the transcript |
createTempShare has NO HTTP creation route, only GET/HEAD on /api/v1/temp-share/[id] |
REFUSED (prints the missing seam instead of faking a link) |
| P0-3 | /compact manual compaction |
updateAgentSessionHistory → compactIfNeeded (needs a real trigger action, not a client-side lie) |
shipped |
| P0-4 | Reasoning surface: per-turn reasoningTokens in the transcript and a /reasoning toggle |
usage.reasoningTokens already recorded by provider streams |
shipped |
| P0-5 | `/queue [list | clear]` follow-up queue for this session | local queue in session state, reuses agentRound |
| P0-6 | `/theme [dark | light | mono |
For /export --markdown, omit --out and --force: Markdown prints in the TUI and can be copied to a file. Direct Markdown file export is unavailable because session content can originate from a remote server. The non-Markdown export retains its existing --out option.
Phase 0 shipped evidence
- P0-1 (
/export): case "/export" in scripts/mso-agent-commands.mjs, tested in scripts/mso-agent-export.test.ts - P0-2 (
/share): case "/share" in scripts/mso-agent-commands.mjs, tested in scripts/mso-agent-export.test.ts - P0-3 (
/compact): case "/compact" in scripts/mso-agent-commands.mjs, tested in lib/agent/session-compact-force.test.ts - P0-4 (
/reasoning): case "/reasoning" in scripts/mso-agent-commands.mjs, tested in scripts/mso-agent-reasoning.test.ts - P0-5 (
/queue): case "/queue" in scripts/mso-agent-commands.mjs, tested in scripts/mso-agent-queue.test.ts - P0-6 (
/theme): case "/theme" in scripts/mso-agent-commands.mjs, tested in scripts/mso-agent-theme.test.ts
Phase 1: medium
| # | Item | Seam | Status |
|---|---|---|---|
| P1-1 | Diff viewer panel from artifact revisions | GET /api/v1/agent-sessions?view=artifact&action_ref=… already returns diff.unifiedDiff |
shipped |
| P1-2 | /stash prompt stash (20 slots, LIFO pop) |
per-process state; pop PRINTS the text because the composer has no prefill seam |
shipped with stated limit |
| P1-3 | /diff [action_ref] binds P1-1 to the last write tool call |
session events | shipped |
| P1-4 | /image <path> attach a local image to the next prompt |
/api/v1/fs/upload, block prepended to the next user message then cleared |
shipped |
| P1-5 | /files read-only file tree of cwd (bounded) |
/api/v1/fs/list |
shipped |
Phase 1 shipped evidence
- P1-1 (Diff viewer panel): renderer in scripts/mso-agent-diff.mjs, tested in scripts/mso-agent-diff.test.ts
- P1-2 (
/stash): case "/stash" in scripts/mso-agent-commands.mjs, tested in scripts/mso-agent-stash.test.ts - P1-3 (
/diff): case "/diff" in scripts/mso-agent-commands.mjs, tested in scripts/mso-agent-diff.test.ts - P1-4 (
/image): case "/image" in scripts/mso-agent-commands.mjs, tested in scripts/mso-agent-image.test.ts - P1-5 (
/files): case "/files" in scripts/mso-agent-commands.mjs, tested in scripts/mso-agent-files.test.ts
Phase 2: large, needs decisions
The renderer half of Phase 2 is planned in detail in TUI-RENDERER-PLAN.md, including why the alternate screen cannot come before a transcript model exists.
| # | Item | Why it is large | Status |
|---|---|---|---|
| P2-1 | Session tabs | Needs alt-screen; without it tabs corrupt scrollback | NOT DONE (deferred) |
| P2-2 | Mouse, panels, scrollbar, split | Requires a retained-mode renderer (OpenTUI/blessed/ink) | NOT DONE (deferred) |
| P2-3 | Configurable keybinds | Keymap loader, conflict detection, ~/.mso/keybinds.json |
DONE (PR #182) |
Deliberately refused
- Inventing reasoning TEXT. The provider streams expose reasoning token counts, not reasoning text. Showing a fake block would be a lie.
- Per-model cost. Not recorded anywhere. The command says so.
- A
/themethat only recolours half the UI. Either everyC.*use goes through the palette, or the command is not shipped.
Gate
Same as PR #165: targeted tests, check-architecture without raising a ratchet,
check-docs, pr-preflight --local, pr-preflight --security, then a draft PR.