Loading...
Loading...
Unified Multi-Agent Harness & Desktop Runtime for AI-Native Engineering.

Pipper is a local developer harness and desktop interface (Omni) that centralizes agent-based coding runtimes. By launching Codex, Claude-Code, and OpenCode subprocesses concurrently within a single sandboxed session, the runtime shares workspace context, synchronizes credential access, and coordinates parallel agent operations. Developers can instruct one agent to implement feature blocks, while other runtimes actively audit outputs, run syntax compilers, and perform local check routines before safe write-backs to the project file system. Powered by Electron and Node, Pipper bridges the gap between fragmented command-line agents and unified IDE workflows.
| Component | Technology / Tool | Description |
|---|---|---|
| Desktop Runtime | Electron, HTML5, Vanilla CSS | Provides a lightweight desktop container that integrates local terminal shells and workspaces. |
| Orchestration Client | Node.js v20+, TypeScript | Runs parallel subprocess instances of Codex, Claude-Code, and OpenCode, managing standard streams. |
| Workspace Sandboxing | Node fs & OS Security Boundaries | Enforces strict read/write boundaries for agent file access, preventing modification of external directories. |
| Credential Gateway | pipper.dev API Service | Centralized authorization service validating model subscriptions and session tokens. |
| Cross-Verification Engine | Llama 3 & Claude Sonnet | Compares agent-generated code drafts, runs automated syntax parsers, and merges refined blocks. |
Minimalist terminal container launching agent sub-processes, embedding shared credentials.
Locally securely binds model authorization tokens for Codex, Claude-Code, and OpenCode interfaces.
Monitors agent syscalls, preventing unauthorized directory traversals or outbound API calls.
Parses abstract syntax trees from different agents to resolve git conflict styles and verify structure.