NotASithLord/peerd
The first AI agent harness native to the browser. A browser extension that runs a full agent loop where you already work: it drives your tabs, spins up sandboxed compute (JS notebooks, WASM Linux VMs, client-side apps), and shares what it builds peer-to-peer. BYOK, no backend, no telemetry.
What it solves
peerd is a browser-native AI agent harness that allows an AI to operate directly within your existing browser tabs and sessions. It eliminates the need for a cloud-based backend or telemetry by running the entire agent loop locally in the browser, giving users full control over their data and API keys.
How it works
The project uses the browser as its runtime and security model, leveraging V8 isolates for sandboxing and WebCrypto for a local encrypted vault. It employs a structural isolation architecture where a main orchestrator agent delegates tasks to separate "actors." These actors run in their own memory heaps (on Chrome) and hold only the tools necessary for their specific environment (such as a single tab, a Linux VM, or a JS notebook), ensuring that the agent holding the API keys never directly touches raw page content or untrusted code.
Who it’s for
Users who want a powerful AI agent capable of driving their browser and running code in sandboxed environments without sacrificing privacy or security, and developers who prefer a "Bring Your Own Key" (BYOK) model with no cloud intermediary.
Highlights
- Browser-Native Runtime: Runs as a Chrome/Firefox extension with no backend or cloud component in the data path.
- Sandboxed Compute: Can spin up JS Notebooks, compiled WebAssembly tools, and full Linux VMs directly in the browser.
- Strict Security Model: Uses a vault for API keys and a delegation system to prevent prompt-injection attacks from accessing sensitive secrets.
- BYOK Support: Compatible with Anthropic, OpenRouter, OpenAI, Z.ai, and local Ollama instances.
- P2P Capabilities: A preview channel includes a WebRTC-based peer-to-peer network for agent-to-agent communication.