KCP Dashboard

Thought Graph Simulated · runs in your browser Live

What am I looking at?

An AI agent is doing a job. To do it well it needs the right internal documents — but you don't want it quietly pulling in anything: stale runbooks, restricted files, paid feeds, or 50 pages that blow the budget.

KCP (Knowledge Context Protocol) sits between the agent and your knowledge and checks every candidate document against a fixed list of rules — gates — before the agent may read it. Pick a scenario below and watch it happen.

🤖 Agent needs knowledge 🚦 KCP runs the gates ✅ only what passes reaches the agent 📊 you see what & why

1 Sessions

Each row on the left is one agent working on one task. The top one is “live”.

2 Steps

The Steps tab is what the agent did. A ✓ means KCP guided that action.

3 Decisions

The Decisions tab is why — which documents were allowed in and which were blocked, and by which rule.

Why does this matter? (30-second version)
  • Cost — every document an agent reads costs tokens (money). KCP caps it and shows where the budget went.
  • Safety — least privilege for knowledge: restricted or out-of-date material is blocked before the agent sees it.
  • Auditability — every decision is deterministic and recorded, so you can answer “why did the agent use that?” months later.
  • Trust — this isn't a guess at the model's hidden “reasoning”; it's the actual, repeatable rule check.
Pick a real-world scenario
1.5×
Real TUI keys — s sessions · g steps⇄decisions · j/k move
Loading scenario…

Active Sessions

One agent, one task per row.
    What the agent did, newest first. ✓ = this action was guided by KCP.