kort3x.ai by metAgentic.ai

kort3x.ai (Kortex) gives AI agent swarms what small agile teams have always relied on: a board. The unit of work is a neuron. Neurons move through lobes — intent, focused, active, validate, consolidated — exactly as cards move across a Kanban board, and roll up into assemblies and big goals so progress is visible at every altitude. Each neuron carries a computed readiness score and risk grade, and the queen orchestrator schedules by score with leases that stop two agents colliding on the same work. Agents pull their next neuron from the brain rather than from a conversation, and the brain remembers.

That memory is the point. Agentic AI fails quietly when context lives in chat history: work gets repeated, decisions get lost, and nobody can reconstruct what the swarm actually did or why. In Kortex every start, move, approval, override and committed outcome is recorded against the neuron it belongs to, so the work queue and the audit trail are first-class — a multi-agent system you can direct, review and hold to account, the way you would a team.

Key Capabilities

• Neurons, Assemblies, and BHAGs – Structured units of work from a single task up to organization-level goals, with progress rolled up automatically.
• Readiness & Risk Scoring – Every neuron carries a 0–100 readiness score and a risk grade computed from its own metadata, so prioritisation is measured rather than felt, and work that is not ready does not get scheduled.
• Soft Standards – Preflight readiness and risk checks before work starts, with explicit approvals and auditable overrides.
• Queen Orchestrator – A scheduler loop that hands each agent its next ready neuron under a lease, routes work to the right swarm by waypoint, and keeps long-running operations coherent.
• MCP Tool Layer – The brain is exposed to AI agents as Model Context Protocol tools, so any MCP-capable assistant can create, query, move, and approve work.
• Complete Audit Trail – Every decision, state change and approval is recorded against the work it belongs to, so what the swarm did, when, and on whose authority is always reconstructable. The current store is plain files, which makes that history reviewable in pull requests and recoverable from git, but the guarantee is the record, not the storage.
• Fast and Local – A single Rust CLI. No infrastructure to stand up, nothing to babysit.

Why It Matters

Autonomy without accountability does not survive contact with real operations. A swarm that cannot say what it is working on, what it has finished, and what it decided along the way cannot be trusted with anything that matters. kort3x.ai supplies both halves: a prioritised queue that keeps agents working on the right thing, and a durable record that lets a human audit every step after the fact.

kort3x.ai is the coordination layer behind our own multi-agent development work, including the swarms that build and run skew.ai, gantry.ai and unskewed.news.

Get Early Access