Dashboard Reframing

We are reframing our entire dashboard around AI agent visibility. Previously centered around traffic egressing to vendors, we will now provide visibility and security for AI agent dataflows.

Core Value Propositions

Discovery / Inventory :

Full runtime visibility into every agent, the MCP servers and tools they call, and the resources and skills they invoke.

Inventory Items:

• Agents ex: claude, cursor, zig
• AI endpoints - LLMS ex : api.anthropic.com
• Users
  • who ran the session
• MCP Servers
• Files
  • Files read/written
• Agent Tools
  • The tools used by an agent
• Models - the LLM model ex: haiku 1.0, opus 4.9
• Endpoints: where agent is running ex: linux server, laptop
• Sessions
  • date / time

The raw material that we mine the inventory items are "Sessions". We know everything possibly knowable about a session. A session will be tagged with inventory items (user, agent, AI endpoint, files, tools, model, etc)

Hidden Risk and Exposure

We not only see the direct calls an AI agent makes, but also any secondary calls. ex: executing a script that accesses github with a local token. We are able to map secondary calls and delegation chains. (ask for clarification if unclear).

Security Policies

We will use the security policies UI / UX to define what is permissable and then to alert on deviations.

TODO :

• Research and discover everything thatt is knowable from an agent session–research this. Assume that we have system priviledged oversight and can see all network requests, file read/writes, MCP sercver requests, etc
• Define the inventory items
• Create good model definition for each inventory item

After we have a good model, we will create a design brief and wireframe–but don't begin those activities yet.