NextClaw reads source material, uses tools, works with local files, and creates documents, charts, code, and apps on a machine you control. Start with one task and learn each part as you need it.
Provide an outcome, source material, and a working directory. The agent can use models, skills, a browser, the terminal, and local files. You can observe, redirect, or stop it, then inspect the actual artifact in the same workspace.
State the material, output format, destination, and boundaries.
It reads files, invokes tools, runs commands, opens the web, or delegates subtasks.
Watch actions, answer questions, add constraints, or stop execution.
Open replies, files, source, Markdown, HTML, images, or a Panel App.
Open local files, Markdown, code, HTML, documents, spreadsheets, and presentations in the session workspace. Inspect HTML as source or rendered output and review code changes as diffs.
Dedicated agents can keep an identity, home directory, memory, skills, and runtime. Package proven work as a skill, connect external tools through MCP, or keep an interactive result as a Panel App.
Scheduled tasks cover briefs, checks, and recurring summaries. Channels let requests and results move through Weixin, Feishu, Telegram, Slack, and other familiar entry points. Prove the work locally before automating it.
