PandaNpc Work lets you give an AI agent work and return to a result you can inspect. Use it for research, organizing material, creating content, or writing and reviewing code. Choose Claude Code, Codex, or PandaCode, then continue from any device.
A workspace for work in progress
Start with a clear outcome: investigate a topic, turn sources into a brief, prepare a draft, or change a repository. The selected agent works in the connected computer and directory you choose. PandaNpc Work keeps the session, approvals, progress, and resulting files or reports within reach, so you can add requirements and review the output.
This is not a promise that every task runs without supervision. Agents act with the tools, permissions, provider account, and working directory available on the connected computer. A machine that is offline, asleep, or disconnected cannot continue the work. Review approvals and results when the task calls for your judgment.
Choose an engine, keep one work surface
Different jobs can call for different engines and model accounts. PandaNpc Work lets you select Claude Code, Codex, or PandaCode without changing how you find the connected computer, open a session, watch its activity, or resume it later. The engine remains responsible for its own provider login, quota, and permission model.
PandaPaw supplies the device bridge and local agent runtime. It connects a Windows, macOS, or Linux computer to your PandaNpc account and runs supported engines there. Your project files and shell stay on that computer; no inbound port is required. Install and manage PandaPaw when you are ready to connect a machine.
Stay involved from any device
Use the web app, desktop app, iOS app, or Chrome side panel to check on a session. Read live output, answer a question, approve a tool action, add a requirement, or reopen work later from another device. Shared connections and read-only transcripts let teammates participate at the level you choose.
PandaNpc Work also connects agents to durable project context when you configure it: Agent Center synchronizes managed guidance in each engine's native layout, notes preserve useful material outside one chat, and the MCP server exposes supported notes, models, and scheduled-work tools to compatible clients.
Project context and knowledge
Context is intentional rather than automatic. Give an agent the project, files, instructions, notes, and connected tools it needs for this task. Agent Center can distribute managed rules, skills, memory, and agent definitions to a bound project; notes provide a knowledge base that compatible agents can read or write. Check the selected engine's configuration and permissions before assuming it can access a source or make a change.
Brand update
Since 2026-09-10, the public product name is PandaNpc Work. Existing PandaNpc documentation and URLs remain in place, and PandaNpcProject remains the engineering repository name. The PandaPaw CLI remains the host bridge and runtime; its command names do not change with the product brand.
Start here
If you already know Claude Code or Codex, open the Quick Start Guide to connect a computer and verify your first agent session. If you want to see the Agent interface before connecting a computer, open PandaNpc Agent in the web app. For provider and model setup, see custom model lists; for common constraints, see the FAQ. For the connected-host workflow, session messages, tool results, and referenced file content can travel through the relay and to the model provider. See the Privacy Policy for the applicable data-handling description.