Secure multi-agent workstation
LINCE turns your terminal into a multi-agent engineering workstation. Run Claude, Codex, Gemini, or any coding agent in parallel — each one sandboxed, monitored, and controllable by voice. No vendor lock-in.
curl -sSL https://lince.sh/install | bash
Linux (fully supported) · macOS (experimental) · Windows (planned)
AI coding agents get full terminal access. Without isolation, one bad command can reach your SSH keys, cloud credentials, or git push to production. LINCE sandboxes every agent with bubblewrap — same tech as Flatpak.
Run 8+ agents in parallel across different projects. The TUI dashboard shows real-time status, token usage, and active tools. Spawn, focus, hide, kill agents with single keystrokes.
Speak and your words are transcribed locally via Whisper, then routed to whichever agent has focus. No audio leaves your machine. Works with any agent, not just one vendor.
Your workflow and security should not depend on a single vendor. LINCE supports Claude Code, Codex, Gemini, OpenCode, Aider, Amp — and any custom agent via config. Switch freely.
A Zellij WASM plugin that acts as your command center. Config-driven — add new agent types with TOML, no code changes.
8+ agents in parallel. Claude, Codex, Gemini, OpenCode, or custom.
Save and restore your agent constellation across sessions.
VoxCode transcriptions piped directly to the focused agent.
Every agent runs inside agent-sandbox. Full autonomy, zero host risk.
Agents auto-grouped by project directory when working across repos.
Add agents via TOML. No code changes, no recompilation.
Interactive wizard: detects your OS, installs prerequisites, sets up sandbox + dashboard. Offers optional VoxCode voice input.
Clone, install sandbox + dashboard, launch. See QUICKSTART.md for details.
Fully supported. Tested on Fedora, Ubuntu, Arch.
Experimental. Sandbox via nono (Seatbelt). Looking for testers.
Planned. WSL2 likely works today. Tracking issue.
You speak ───> VoxCode ───> Whisper transcribes locally │ Dashboard <─── routes to focused agent │ agent-sandbox <─── isolates execution ┌──────┼──────┐ Claude Codex Gemini ...any agent │ │ │ project/ project/ project/ (sandboxed)
Sandbox + Dashboard + Hooks. The multi-agent orchestration platform.
Voice input for AI agents. Local Whisper transcription, Zellij pipe integration.
Text-to-Speech with local GPU/CPU engines. Kokoro and Piper TTS.
Assess any project's readiness for agentic coding. agentskills.io compliant.