🛡️ What we’re building: KAP (Kernel Assurance Plane)
The enforcement floor of the ArmorIQ stack. KAP closes the gap between authority refinement and OS execution where individually-legal syscalls add up to something the task never authorized. “Refactor the config” becomes
fork → execve(curl) → connect(external) → send(config) every step legal, the aggregate an exfiltration. Linux tells you a syscall is permitted; it can’t tell you the authority behind it still matches the task. KAP enforces one invariant: an operation runs only if it’s derivable from committed lineage and never exceeds the authority running it. Authority can evolve but only through committed, refinement-bounded lineage, never silent drift. Dropping soon.
🧠 PAP (Purpose Assurance Plane): the dashboard is taking shape
We’ve got a live build of the PAP control plane running. Still in dev and testing, but it’s real. PAP governs the most underprotected surface in agent systems: plan refinement, the space between intent and execution where authority quietly expands at every step. “Summarize customer churn” becomes “query DB, export, external API,” each step locally valid, the aggregate a leak. The dashboard makes the invariant visible. You watch the Cone of Intent narrow as uncertainty drops, while authority stays inside a bounded lattice where interfaces only shrink and constraints only grow. Plans that add a new tool or weaken a constraint get filtered before they reach execution. The dangerous moment isn’t execution, it’s plan construction, and now you can see it governed in real time. Rough edges, more to come.
🛠️ ArmorTools: one dashboard for every agent you run
We’re building out ArmorTools, the unified console for the whole ArmorIQ stack. Claude Code, Codex, OpenClaw, GitHub Copilot, all your agents under one roof, with intent security enforced on every tool call. One view shows total events, violations, allow rates, and critical findings across ArmorClaude, ArmorCodex, ArmorCopilot, and ArmorClaw, so you can see exactly where authority is being exercised and where it’s being blocked. Activity heatmaps, streaks, and a live feed of the last events across every agent. The point isn’t just observability, it’s that each of those events was checked against committed intent before it ran. Out soon.
