Bright Vision

Website: bright-vision.digitaldefiance.org

A lightweight, cross-platform desktop IDE built with Tauri and React — local LLM first, spec-driven tasks, and superproject/submodule git — powered by Bright Vision Core (a fork² on cecli: bundled cecli + ported headless HTTP API from our earlier aider-vision-core — see lineage).

🚀 Features

• Vision API: All prompting goes through the HTTP API (React is the head; core is headless under bright-vision-core/).
• Local LLM: Ollama + built-in Local LLM panel; defaults to on-device models — see docs/LOCAL_LLM.md.
• Chat UX: Streaming replies, Thinking/Answer sections, proposed-edit accordions, confirm flow, queue/stop, /add path completion (desktop), image/PDF attach.
• Tasks & specs: EARS/spec-driven workflow (.bright-vision/todos.json, generate/refine spec, steered Implement steps) — docs/SPEC_DRIVEN_DEV.md.
• Git tab: Working tree, diffs, commit graph, stage/undo, auto-stage after turns (desktop).
• Superproject + submodules: Multi-repo workspace via RepoSet — docs/SUBMODULE_VERIFICATION.md.
• Native performance: Rust + Tauri v2 on macOS, Linux, and Windows.

🗺 Roadmap status

Living backlog: docs/ROADMAP.md · project site.

Current focus: cecli engine migration (default engine bright-vision-core) then dogfooding on real repos (yarn tauri dev).

Quick checks: yarn test:local · core: yarn test:bright-core

🛠 Tech Stack

• Backend: Rust + Tauri v2
• Frontend: React + TypeScript + Vite
• Engine: Python — bright_vision_core (HTTP/SSE) on cecli (coders, agents, commands)
• Styling: MUI v6 + Emotion; global SCSS in src/styles/
• Package Manager: Yarn

🔒 Local LLM first (privacy-first)

Bright Vision targets local inference on your hardware:

1. Install Ollama.
2. Copy local-llm.env.example → local-llm.env (DATA_MODEL, optional OLLAMA_HOST) — docs/LOCAL_LLM.md. Local LLM is built into the app (Rust + Python); no local-llm.sh required.
3. Desktop: Terminal → Local LLM → Start, or Auto before session + Terminal → Start.
4. Chat when the session is live.

Cloud providers still work via LiteLLM model strings and API keys in the environment.

Note

Dogfooding target: Bright Vision builds itself. Primary local setup: Qwen Coder 3.6 27b q4_K_M on Apple Silicon (64GB RAM).

📦 Getting Started

macOS (Homebrew)

brew tap digital-defiance/tap
brew install bright-vision

Installs Bright Vision.app to /Applications/. Tap: digital-defiance/homebrew-tap

Legacy cask name aider-vision may still exist during transition.

From source

Prerequisites

• Node.js (v18+)
• Rust (latest stable)
• Yarn (v3+)
• Python 3.10+ (for the engine submodule)

Installation

git clone https://github.com/Digital-Defiance/bright-vision.git
cd bright-vision
git submodule update --init --recursive
yarn install
source activate.sh    # editable bright-vision-core + uvicorn
yarn tauri dev

Production build: yarn tauri build

Legacy engine: BRIGHT_VISION_ENGINE=bright-vision-core source activate.sh

⚙️ Configuration

Settings → Model & system: LLM model (default ollama_chat/qwen3.6:27b-q4_K_M), LiteLLM extra params, project workspace, context files, engine path (bright-vision-core).

Settings → Appearance: UI / chat / terminal fonts (default chat: Glass TTY VT220).

See docs/DEVELOPMENT.md, docs/USER_WORKFLOW.md, docs/LOCAL_LLM.md.

📚 Documentation

Doc	Topic
CECLI_MIGRATION_ROADMAP.md	Engine port status (cecli + bright_vision_core)
LOCAL_LLM.md	Ollama, Local LLM panel, local-llm.env
ROADMAP.md	Product backlog
TESTING.md	yarn test:local / test:full
DEVELOPMENT.md	Setup and conventions
IPC.md	HTTP API and SSE events
TROUBLESHOOTING.md	Connecting, orphaned :8741
SUBMODULE_VERIFICATION.md	Superproject + submodules
SPEC_DRIVEN_DEV.md	Tasks / spec-driven workflow
RELEASE.md	Tag core and bump submodule

📜 License

MIT — see LICENSE.

Copyright (c) 2026 Digital Defiance, Jessica Mulein

🤝 Contributing

Read docs/ROADMAP.md and docs/CECLI_MIGRATION_ROADMAP.md before substantive work. Open issues with repro (workspace path, expected vs actual).