BrightVision is a local-LLM-first desktop IDE (Tauri + React) with a headless Python engine (Cecli on cecli). This page is the product catalog; tactical status lives in ROADMAP.md.
| Area | What you get |
|---|---|
| Left rail | Chat, Tasks, Terminal, Git, Editor, Settings — full-surface tabs, not a VS Code clone |
| Activity bar | Live phase (connecting, scanning repo, waiting on model), optional Response / Think timers, ~ETA when history exists |
| Resource overlay | CPU / RAM / GPU % in the rail (desktop; system-wide polls) |
| Session chip | File count + token context estimate; click for files_in_chat list |
| Feature | Notes |
|---|---|
| Streaming | Thinking / Reasoning / Answer sections, section duration chips |
| Code fences | Syntax-highlighted blocks + Mermaid diagrams (not plain-text dumps) |
| Proposed edits | SEARCH/REPLACE accordions; “applied” labels after done |
| Confirm flow | Core prompts with Yes/No; auto-approve budget in Settings |
| Queue & Stop | Follow-up messages while busy; clear queue from header |
| Clear history | Sticky ✕ above transcript; confirms; sends /clear when session is live |
| Slash commands | / palette + quick chips; /add Tab path complete (desktop) |
| Agents | /agent, /invoke-agent, /spawn-agent, /reap-agent chips; sub-agent registry chips |
| Attach | Images/PDF, terminal tail, folder (desktop picker / web path) |
| Suggested files | Parses assistant Answer for paths; tray with Add all / proceed |
| Empty LLM | Local-friendly copy + Retry / Retry with hint |
| Ollama client cmds | /ps, /tags, /models — tables in chat (not sent to core) |
| Model router | Fast vs heavy tier; escalate offer; force tier chips |
| Token footer | Usage line after turns (legacy Tokens: + cecli ↑↓) |
.cecli/todos.json+ workspace HTTP API- Layered specs (requirements / design / tasks),
depends_on, templates - Generate / Refine spec, background jobs
- Implement steered steps with active task chip
- Branch, ahead/behind, working tree
- Stage / commit / undo (desktop Rust git)
- Commit graph
- Auto-stage policy after turns (Settings)
- File tabs + CodeMirror 6
- Collapsible explorer with git status badges
- Open from chat / suggested files
- Optional language packs (Settings)
- Start / Stop Cecli (spawns API on desktop)
- Technical log (stderr + debug) vs user-facing chat
- Local LLM panel — Ollama tags, load model, ping
- Ping stack in Settings — Ollama 1-token probe + Vision API
/health
- Model, workspace, context files, engine path, appearance fonts
- Thinking timers + timing history (avg/peak CPU/RAM/GPU, TPS, CSV export)
- Suggested-files prefs (auto-add, auto-proceed)
- Model hopper / router (UI or
local-llm.envFAST_MODEL/HEAVY_MODEL/MODEL_ROUTER) - Resource overlay toggles
- Agents — sub-agent registry + cecli doc links
- macOS Homebrew
brightvision, DMG build docs - Web dev — Vite proxy to core on
:8741 - Tests — Vitest, Playwright e2e, core pytest (TESTING.md)
- Full TUI sub-agent pills / agent switcher in the header
- LSP / extension marketplace
- GPU-driven ETA (logged for research only)
Wired in README.md (GitHub user-attachments, May 2026):
| Capture | URL id |
|---|---|
| Chat + Mermaid | 09652221-d4bc-4dc3-8f25-56529d04b015 |
| Activity bar + confirm | 21bfe84c-c134-43c6-82c6-71894133b42e |
| Tasks | f9ea3523-aa0d-405c-aff8-371d59412c36 |
| Editor | 3e21ed90-362d-49f3-93dd-f0a16d786866 |
| Git | ead7a30a-b01b-48ff-be71-e15689eeeba7 |
| Settings timing + overlay | dc6a8e0c-302e-4d7a-b455-7e1c7b7f28a0 |
Still useful for a future pass: Agents chip row, proposed-edit accordion close-up.