Stop juggling single-model chat windows. Run Claude, OpenAI, Kimi, and Ollama in a real-time multiplayer huddle on your Mac with 100% local-first privacy.
Click between seats to see how multi-agent reviews unfold live in real-time.
A desktop huddle environment built for real-world software development.
4-stage huddle engine (Proposal ➔ Peer Review ➔ Response ➔ Refined Plan) where agents defend, concede, and refine architectural blueprints together.
Consolidated design plans automatically sync to brain/my_plans.json and workspace configs (.cursorrules) in the background without manual buttons.
Agents instantly search across all past design reviews, historical architectural decisions, and chat transcripts right on your Mac's hard drive.
All API keys are encrypted via macOS native Keychain using Rust's keyring crate. Zero plaintext credentials on disk.
Pair cloud reasoning (Claude/GPT-4) with 100% offline local Ollama models (Qwen, Llama 3, DeepSeek R1) so confidential code never leaves your Mac.
Canonical path containment ensures agent file operations remain strictly inside your active workspace directory.
How Kaldi Station compares to standard cloud chatbot tools.
| Feature / Dimension | ⚡ Kaldi Station | ChatGPT (Web) | GitHub Copilot |
|---|---|---|---|
| Data at Rest | ✓ 100% Local on Your Drive | ✕ Cloud Database | ✕ Cloud Database |
| API Key Storage | ✓ Encrypted OS Keychain | ✕ Account Login | ✕ Account Login |
| Multi-Model Huddles | ✓ Parallel Multi-Vendor | ✕ Single Model | ✕ Single Model |
| Offline Model Support | ✓ Air-Gapped Ollama | ✕ Cloud Only | ✕ Cloud Only |
| Middleman Server Telemetry | ✓ Zero (Direct HTTPS) | ✕ Central Server | ✕ Central Server |
Download the signed standalone macOS installer package.
Drag your downloaded KaldiStation.dmg file into the drop-zone below to automatically verify its cryptographic integrity on-device.