netclaw.
Explore the preview
Every great network starts with a good question.
Bring your tools, procedures, and network context into one workspace. Ask in Chat. Follow the work. Open the evidence beside the answer.
Find your workflowA desktop in development. A bigger way to work.Follow a question about SPE1 into the Inspector, ask about selected evidence, and explore the tools that shape your workspace.
4 min 24 sec · Narrated feature tour of the current desktop and Sandbox Lab workflow. Includes Chat, Inspector, Skills, Steering, Risk, Automation, and a look at Change control. The recording uses Lab mode; the Change control segment is a feature overview.
Start in Chat.
Build in Projects.
Stay in control.
This desktop app explores a new experience inspired by John’s NetClaw. Discover the original project, explore John’s work at Automate Your Network, and connect with the community at VibeOps Forum.
A new agent bar puts your tools, skills, and standing instructions within reach. Chat keeps the work in view, with the evidence one click away.
Manage MCP servers in a dedicated panel. Review connection states, tools, and environment bindings. Add servers from the NetClaw catalog or edit your configuration.
Browse and enable skills. Keep standing instructions in Steering, with always-on guidance or a #mention for a specific conversation. Guidance never adds permissions.
A compact composer leaves room for the conversation. Open tool activity in Details, read saved command snapshots, and ask about a selected excerpt in the Inspector.
Development preview. A connected MCP server is not proof of device reachability. Network environments and execution hosts are separate, explicitly configured choices.
Explore how a switch replacement can move from a question to a project and a reviewed change. Start wherever your work starts.
Illustrative workflow, not a live run. Chat, bounded project planning, and explicitly enabled environment-scoped MCP tools are implemented. Local lab schedules and member routing are implemented. Production writes have separate change-control checks and exact operator approval; live production acceptance remains unverified.
A quiet Chat surface. Space for project briefs. An agent bar for MCP servers, skills, steering, and change control. An Inspector for the evidence behind the answer.
Start with a question or a starter card. The compact composer keeps your provider, model, and network target together. This is a rendered UI preview with no model connected.
The local Core stores chat history and each conversation’s own draft. Start a new chat, then return to the question you left unfinished.
Choose a provider and model, target a network environment, and follow attributed tool results. Reported context capacity and measured prompt usage stay visible.
System, Light, and Dark appearance. Coral accents, a slimmer sidebar, and dedicated agent controls. A sample case when you want to explore.
The Border is the conversational entry point. Choose local Ollama or a supported cloud connection. Enabled network tools stay scoped to the environment you explicitly select.
Configure local specialist Claws with their own models and scoped capabilities. The Border routes bounded tasks to eligible members and keeps returned artifacts attributed. External peering remains planned.
Tools and credentials remain scoped to each member. Lab evidence, a model switch, or a delegated task never grants production approval.
The destination is ambitious.
Here’s where the work stands today.
A native desktop backed by a separate local Core.
Real model and read-only lab flows have been exercised. Each integration still needs its own verification.
Build on verified evidence while keeping authority explicit.
Development preview · Open-source, self-hosted direction
