PANDA Framework

open-source framework for Autonomy

PANDA Framework

PANDA is an open-source framework that provides the structure and foundation for AI Agents, organized into Perception, Analysis, Network, Decision, and Action.

View on GitHub Get Started
Perception - continuously acquires information from the environment and internal sources.
Analysis — standardizes, classifies, evaluates, and predicts information before sharing it with the Network.
Network — Exchanges context and knowledge to expand available options with network for decision options.
Decision — Evaluates decisions, possible outcomes and goals proceeds to select next action(s).
Action — Executes the selected actions through available capabilities.

Built Around Observations And Actions

PANDA treats every input as an observation and every output as an action.

Modular

Each agent state can be implemented, extended, or replaced without forcing the whole runtime into a fixed loop.

Observable

Agent behavior should be inspectable and traceable through observation history, state transitions, and dispatched actions.

Connector-first

Filesystems, browsers, APIs, sensors, queues, databases, and other systems use one connector shape.

Repository Scaffold

The project is organized as a lightweight pnpm monorepo with apps, shared runtime packages, examples, and documentation.

  • apps/cli
    Command surface for PANDA.
  • apps/daemon
    Local Fastify daemon with HTTP API and WebSocket events.
  • apps/dashboard
    React, Vite, TypeScript, and Tailwind dashboard.
  • packages/core
    Observation bus, scheduler, action dispatcher, connectors, memory, sessions, and config.
  • packages/sdk
    Typed daemon client and public observation/action types.
  • packages/shared
    Shared schemas, IDs, timestamps, and logger utilities.

Development

Install dependencies, run the daemon and dashboard together, or build and test the full workspace.

pnpm install pnpm dev pnpm build pnpm test