Perci is an AI workspace with open-source roots — a desktop application that brings together multi-provider AI chat (including OpenClaw and Hermes), agent orchestration, deep research, and a code workbench under one roof. Forked from Open Claude, it has evolved into something substantially more: a general-purpose tool for anyone who works with AI, writes code, or needs to get real things done.
This is not a chat app. It's a command center.
Switch between OpenAI, Google Gemini, Anthropic Claude, Groq, DeepSeek, or local models (Ollama, LM Studio, vLLM) — all in one interface. Each provider brings its own strengths; Perci lets you use the right tool for the job.
An autonomous research agent that plans searches, evaluates sources, synthesizes findings, and produces formal reports with abstracts, methodology, and citations. This isn't prompt-and-pray — it's an iterative loop that keeps digging until the question is answered.
- Code Mode: A dedicated code editor surface for working with AI on implementation tasks
- Cowork Mode: Run multiple coding agents (Claude Code, Aider, Codex, Cursor CLI, Copilot, Antigravity) in parallel, each in its own workspace, coordinated through Perci's mission control system
Orchestrate complex multi-step work. Launch agents, track their progress, validate outputs, and manage dependencies — visualized through a transit map that shows every run's state transitions. This is the operational backbone that makes Perci more than a single-chat interface.
AI-generated code (HTML, React, SVG) doesn't just appear in a chat bubble. It opens in a dedicated side-by-side panel with live preview, so you can iterate on working software, not just read about it.
Perci includes native support for OpenClaw — a production-grade local AI gateway. This gives you persistent agent memory, cross-session continuity, and access to local models through a unified interface. Perci auto-detects OpenClaw on your system and installs the gateway if missing.
Perci's Mission Control orchestrates agent workflows, visualizes complex tasks, and provides detailed insights into every step.
This view captures a live run of an "OpenClaw integration health" mission. You can see the agent's progress through individual checkpoints, the commands executed, and the final report.
The Transit Map provides a high-level overview of agent missions. Each node represents a distinct step or dependency, allowing you to quickly understand the flow and state of complex tasks.
A built-in terminal for running shell commands, managing dev servers, and interacting with your system — all without leaving the workspace.
Watch models think in real-time. Collapsible reasoning sections show the AI's thought process, token usage, and duration — streamed as it happens.
- Node.js 18+
- npm or yarn
- Git
git clone https://github.com/toshon-jennings/perci.git
cd perci
npm install
npm run devFor the desktop app (Electron):
npm run electron:devTo package for your platform:
npm run electron:build # default platform
npm run electron:build:mac # macOS
npm run electron:build:win # Windows
npm run electron:build:linux # LinuxPerci requires API keys for cloud providers. Keys are stored locally in the Electron app-data file (encrypted via OS safeStorage) — never sent to any server. Add them in Settings → API Keys.
Local models (Ollama, LM Studio, vLLM) require no API key.
src/
├── components/ # React UI components
│ ├── AgentsPanel.jsx # Multi-agent orchestration UI
│ ├── ArtifactPanel.jsx # Code/preview panel
│ ├── BuildMode.jsx # AI app builder
│ ├── CodeMode.jsx # Code editor surface
│ ├── CoworkMode.jsx # Parallel agent sessions
│ ├── MissionControl.jsx # Mission orchestration dashboard
│ ├── ModeSwitcher.jsx # Chat / Build / Code / Cowork
│ ├── Terminal.jsx # Built-in terminal
│ └── ...
├── context/ # Global state (ModeContext, ChatContext, BuildContext)
├── lib/ # Core logic
│ ├── llm/ # LLM provider clients (factory pattern)
│ ├── IntelligentSearchTool.js # Deep research & web search
│ ├── missionControl.js # Mission run tracking & validation
│ ├── terminalBridge.js # Terminal IPC
│ └── ...
electron/ # Electron main process
- Frontend: React 18, Tailwind CSS, Framer Motion
- Build: Vite 5
- Desktop: Electron 39
- Markdown: react-markdown + remark-gfm
- Syntax Highlighting: react-syntax-highlighter
- Icons: Lucide React
npm run dev # Start dev server
npm run build # Production build
npm run preview # Preview production build
npm run lint # Lint
npm run electron:dev # Electron dev mode
npm run electron:build # Package desktop appPerci's codebase is organized around a central event bus (appendMissionRunEvent()) that connects the mission control system to the UI layer. The graph analysis (514 nodes, 1047 edges, 31 communities) shows clean separation of concerns with zero import cycles.
Key architectural communities:
- Mission Event Logging — Central event bus (19 edges, #1 hub)
- App Core and Providers — Window/dock system, mode routing (#2 hub)
- UI Navigation and Settings — Mode switcher, settings, API key management
- LLM Provider Clients — Factory pattern across 6+ providers
- Code and Cowork Modes — Parallel agent orchestration
- Artifact Preview Security — Sandboxed iframes, CSP, budget enforcement
See docs/architecture/GRAPH_ANALYSIS_2026-06-07.md for the full breakdown.
- Open-claude by Damien Chakma — the original fork source. Perci's core chat, thinking UI, and artifact system started here.
- OpenClaw — the local AI gateway that Perci integrates with for persistent agent memory, cross-session continuity, and local model access. Perci shells out to the
openclawCLI; the dashboard and agent bridge are custom-built on top.
MIT — see LICENSE.