Asking AI about a network
View ▸ Ask AI panel (Ctrl+Shift+A) answers questions about the open project: ready-made ones such as Explain this network, What the platforms add, Clean up the asset register, Models, firmware and lifecycle and Questions for the previous owner, or your own. It is optional and does nothing until you set it up.
On your Claude plan
Anthropic does not allow other applications to sign in to a Claude account, so SubnetSleuth never does. It works through Anthropic's own apps instead, under your own sign-in (single sign-on included):
- Claude Code: choose Claude Code in Ask with and Sign in with Claude. Questions run through your own Claude Code.
- Claude Desktop and Cowork: Connections ▸ Use your Claude plan ▸ Add to Claude Desktop, then restart Claude Desktop. In a chat or in Cowork, Claude can read the project you last opened or saved in SubnetSleuth, read-only. The desktop extension does the same in one click, and your organisation can allow-list it.
- claude.ai by hand: Briefing for a chat saves one file to attach and copies the prompt; Data folder for Cowork saves the full export for Cowork to read. Nothing is sent by SubnetSleuth.
Through an API
Save an AI provider under Connections ▸ AI providers: an API key is an ordinary saved credential, and the cloud sign-ins are each cloud's own, so no password is stored.
| Provider | Signs in with |
|---|---|
| Claude API (Anthropic) | API key, or an Anthropic Console sign-in |
| Claude Platform on AWS, Amazon Bedrock | AWS profile or single sign-on, or a key |
| Google Cloud Vertex AI | Google Cloud sign-in, or a service account key file |
| Microsoft Foundry, Azure OpenAI | Sign in with Microsoft (Entra ID), the Azure CLI's account, or a key |
| OpenAI, Google Gemini, Mistral, xAI, OpenRouter | API key |
| Ollama, OpenAI-compatible gateways | None, or a key if the gateway needs one |
Each provider offers its current models and effort levels; Refresh asks the vendor which models your account can use, and Test does a one-word round trip.
What is sent
- Quick questions send a summary of the project. Thorough questions let the model list, read and search an export of the project, and nothing else.
- Show what will be sent previews it exactly.
- Masking replaces addresses, MACs and serials (or names too) with tokens before anything leaves your computer, and puts them back in the answer.
- Questions go only to the provider's own address over HTTPS.
Answers can be kept with the project or saved as Markdown.