Kucedr is a conversational desktop assistant that can take actions through tools instead of only returning text.
Chat
The assistant streams GitHub-flavored Markdown, groups tool activity into expandable status summaries, displays generated media inline, and accepts file attachments. You can stop an active run, copy or reply to a message, and read assistant responses aloud. The composer does not check attachment type, size, count, or model support before sending; errors may appear while the request runs.
Sessions are stored locally and restored when selected. The chat view loads the most recent stored messages for the active session.
Built-In Tools
The agent can work with:
- Files and structured multi-file patches.
- Shell commands and retained processes.
- Brave or Tavily web search and public-page fetching.
- Managed browser automation.
- Image, video, music, and sound-effect generation.
- Workspace profile files. Saved conversations are processed by a separate memory module when enabled.
- Skills and dynamically loaded MCP tools.
- Schedule records and health-check configuration.
Subagents
The main agent can run one independent subagent with a fresh conversation and the same tool set, except for further subagent spawning. Subagents are non-interactive, so any action whose policy resolves to Ask is denied.
Host elevation, gateway execution, and remote-node execution are not implemented in the current runtime.
Continue with Browser Automation or learn how Memory And Workspaces provide persistent context.