The agent-of-agents. Routing, memory, policy, settlement, reputation, self-learning.
Prime doesn't do work. Prime decides who does work, under what authority, with what memory, paid how much, verified how.
Every action: sign → route → log → settle → attest. Zero implicit trust.
┌─────────────── Kynetra Prime ───────────────┐
│ Identity & Delegation (kynetraauth) │
│ Skill Routing (registry) │
│ Memory Governance (CMS + Books) │
│ Policy Engine (OPA-style) │
│ Payment + Settlement (AMP) │
│ Reputation Ledger (L0 attestations) │
│ Self-Learning (Thanos + SGAI) │
│ Audit (Vigil) │
│ Multimodal (Vectron, Visual)│
└─────────────────────────────────────────────┘
spawn → authorise → plan → delegate → execute → settle → attest → learn → retire
Every step signed by principal, routed through Policy, logged by Vigil, settled via AMP.
agent {
id: DID // decentralized identifier
principal: DID // who owns this agent
skills: [SkillRef] // capabilities
wallet: AMPAccount // payment escrow
memory: MemoryScope // read/write boundaries
policy: PolicyRef // auth rules
reputation: Score // track record
cost_cap: Money // max budget
}
Three scopes, all signed and auditable:
Every read logged. Consent enforced at L2.
OPA-style rules: jurisdiction, budget, data class, agent reputation floor. Violations fail closed.
Thanos monitors outcomes; synthesises new skills from successful delegations. Self-Growing spawns specialist sub-agents when a skill gap is detected.
v0.1: Live in production (HB Ventures 32-app portfolio)
Modules: Spread across `/Users/kr/Developer/kynetra-*/`