A native desktop AI workspace — multi-model chat, boardroom consensus, head-to-head evaluation, and live diagnostics — all running locally on your hardware.
OpenMind is a Tauri 2 desktop application that gives you a rich, performant AI interface for locally-hosted models served by Ollama. No cloud dependency, no API keys, no data leaving your machine.
It goes beyond a simple chat window: multiple models can deliberate together in Boardroom mode, compete head-to-head in Dojo mode, and a live Diagnostics dashboard surfaces server health, system telemetry, model inventory, and shell-script integration — all in one native window.
- Streaming token-by-token output with a live cursor
- Full conversation history persisted to
localStorage <think>reasoning blocks rendered as collapsible sections with preview text- Auto-resizing textarea with character ÷ 4 token estimate
- Stop mid-stream without losing partial output
- Accordion dropdown showing all available Ollama models
- Per-model disk size chip pulled from
/api/tags - Drag a model row directly onto the Dojo tab to add it as a contestant
- One-click floating chat window per model — compare responses side-by-side
Six injected agent roles (advocate, critic, analyst, devil's advocate, expert, generalist) deliberate on the same prompt simultaneously, then a configurable synthesizer model produces a structured consensus report.
- 4 preset boards: Classic Triad, Devil's Court, Full Board, Peer Review
- 5 built-in debate / analysis prompt presets
- All advisor streams run in parallel — no waiting for turn-taking
- Re-synthesize with a different model without re-querying advisors
- Stop All aborts every stream simultaneously
An evaluation harness where N models compete on the same prompt, judged by a configurable model on user-selected rubric criteria.
- Rubric: accuracy, conciseness, reasoning, code correctness (toggle per criterion)
- Blind mode hides model names from the judge to eliminate name bias
- Running scoreboard with cumulative wins + average score, persisted across sessions
- 5 built-in test prompts (code challenge, logic puzzle, concept explanation, debug challenge, debate)
- Auto-fit grid layout — panels resize as models are added / removed
- Live CPU usage bar with colour-coded thresholds (green → amber → red)
- Network receive / transmit rates from the Rust
sysinfobackend - Endpoint tester: probe
/api/tags,/api/version,/v1/modelswith raw response viewer - Chat inference test: non-streaming end-to-end roundtrip with latency
- Running models list with VRAM usage from
/api/ps - Shell Scripts card — run
test-opencode-ollama.sh,curlllama.sh,sync-ollama-models.shfrom the UI, output displayed in a terminal-style panel with ANSI stripping - Persistent connection log that survives tab switches
- Pulsing pill in the header: green ring (online), spinning ring (checking), solid red (offline)
- Latency badge updated every 15 s
- Click to jump to Diagnostics
openmind/
├── src/ # React 19 / TypeScript frontend
│ ├── App.tsx # Root — state, polling, layout
│ ├── components/
│ │ ├── NavBar.tsx/.css # Styled tab navigation (glyph + label + desc)
│ │ ├── ConnectionStatus.tsx/.css
│ │ ├── DiagnosticsPage.tsx # Full diagnostics dashboard
│ │ ├── ModelPicker.tsx/.css
│ │ ├── FloatingChat.tsx/.css
│ │ ├── MarkdownMessage.tsx # react-markdown + think-block parser
│ │ └── …
│ ├── boardroom/
│ │ ├── BoardroomPage.tsx # Orchestrator: fan-out + synthesis
│ │ ├── boardroom-client.ts # streamAgentResponse / streamConsensus
│ │ ├── boardroom-types.ts # AgentRole, ROLES, PRESET_BOARDS, …
│ │ └── components/ # AgentPanel, ConsensusPanel, BoardroomInput
│ ├── dojo/
│ │ ├── DojoPage.tsx # N-model grid + judge + scoreboard
│ │ ├── dojo-client.ts # streamContestant / streamJudge
│ │ └── dojo-types.ts # DojoRubric, JudgeResult, …
│ └── lib/
│ └── ollama-client.ts # All Ollama HTTP: chat stream, tags, ps, status
└── src-tauri/
├── src/lib.rs # get_system_stats, run_diagnostic_script, shutdown
└── Cargo.toml # sysinfo + tauri + serde
| Pattern | Where used |
|---|---|
useRef accumulator + functional setState | Streaming fan-out in Boardroom and Dojo — avoids stale closures |
Module-level let _persistedLog | DiagnosticsPage log survives React remount (tab switch) |
AbortController per stream | Every stream independently stoppable; stopAll() aborts the entire set |
CSS custom property --tab-color / --role-color | NavBar tabs and BoardroomPage agent panels take per-entity colour from data |
localStorage persistence | Chat history · selected model · Dojo scoreboard |
Drag-and-drop via HTML5 dataTransfer | Model rows → Dojo tab; carries model name string |
| Layer | Technology |
|---|---|
| Desktop shell | Tauri 2 |
| UI framework | React 19 with TypeScript strict mode |
| Build tool | Vite 7 (dev server port 1420) |
| Rust runtime | sysinfo · serde · tauri-plugin-opener |
| AI backend | Ollama — /api/chat · /api/tags · /api/ps |
| Markdown | react-markdown · remark-gfm · rehype-highlight |
| Package manager | pnpm |
- Node.js 18+ and pnpm (
npm i -g pnpm) - Rust 1.75+ — install via rustup
- Tauri CLI —
cargo install tauri-cli - Ollama running and reachable (default:
http://10.0.0.155:18080)
git clone https://github.com/openmindx/openmind.git
cd openmind
pnpm install
pnpm tauri dev # opens native window with HMRpnpm dev # http://localhost:1420pnpm tauri build # creates installer in src-tauri/target/release/bundle/Three shell utilities live in diagnostics/ and can be run from the Diagnostics tab in the UI or directly from the terminal:
# End-to-end OpenCode + Ollama integration test
./diagnostics/test-opencode-ollama.sh
# LAN discovery, connectivity verification, curl command map
./diagnostics/curlllama.sh
# Write available Ollama models to OpenCode config
./diagnostics/sync-ollama-models.shEdit src/lib/ollama-client.ts → defaultConfig:
exportconstdefaultConfig={ollamaUrl: 'http://10.0.0.155:18080',// change to your Ollama hostmodel: 'qwen3-coder:30b',// default model};A settings panel (edit URL/port without touching source) is on the roadmap.
See docs/ROADMAP.md for the full planned feature list.
Selected upcoming items:
- Settings modal — configure Ollama URL + port in-app
- Message editing and response regeneration
- Multiple named conversation sessions
- In-app model browser and pull / delete
flexlayout-reactmulti-panel workspace- Slash commands —
/model,/boardroom,/dojo,/clear,/export - N-way Dojo tournament bracket with round-robin view
- Boardroom vote display — which advisors agreed / diverged
| Document | Description |
|---|---|
| ROADMAP.md | Feature backlog organised by area |
| BOARDROOM.md | Boardroom mode architecture and design notes |
| DOJO.md | Dojo evaluation harness documentation |
| INDEX.md | Documentation index |
MIT License — see LICENSE for details.
© 2026 Professor Codephreak · Built with Tauri · React · Ollama