Index

Documentation for Index

Fleet Intelligence — 14 Sovereign Agents

The HLD Fleet is a cluster of 14 purpose-built AI agents operating as server-native workers with direct machine access. Each agent maintains sovereign state (Ed25519 key, skill registry, tool access) while sharing the Fleet intelligence layer.

Agent Roster

AgentDesignationDomainPrimary Skills
cipherGeneral of CodeInfra Architecture, SecurityDependency audits, JWT flows, build pipelines, vulnerability research
axonMulti-Agent OrchestratorOrchestration, Task DecompositionParallel dispatch, workflow supervision, capability matching, progress tracking
prismaCreative LeadUI, Visualization, ThemingLanding pages, Tailwind components, Chart.js dashboards, dark-mode theming
pulseOperations LeadDevOps, DeploymentNginx TLS, Docker deployment, VPS cleanup, log rotation, service health
thetaDeep ResearchLiterature, Benchmarks, EvaluationModel evaluation, technical reconciliation, benchmark design, paper synthesis
vectorData AnalystSQL, Sports Modeling, RiskPostgreSQL pipelines, sports odds modeling, risk dashboards, Polymarket bridges
cyChief of StaffCommitments, OperationsWeekly reset, inbox triage, document-to-actions, commitment tracking
gechoBridge GatewayGo Infrastructure, InteropBroker services, Gechos between Fleet and external systems
eveAgent AuthoringScaffold, Lifecycle, SkillsDurable agent generation, tool registration, skill hot-loading, version pinning
novaInnovationPrototypes, Spikes, IncubationExploratory prototypes, spike validation, new capability incubation
titanInfra ArchitectBlockchain, ICP, Sovereign PatternsDigital immortality contracts, ICP integration, sovereign backend patterns
fluxStream EngineerReal-time, WebSocket, BettingLive data pipelines, betting advisory slates, WebSocket orchestration
zenithPerformanceLatency, Quantization, GPUInference benchmarking, GPU tuning, quantized serving, throughput optimization
whaleDeep DiveDatasets, Model SurgeryHuggingFace pipelines, dataset orchestration, model surgery, anti-gravity integration

Dual-Mode Dispatch

1. Sovereign Routing (Direct)

POST /cipher/run
Authorization: Bearer <id_token>
{ "objective": "Audit Ed25519 nonce verification in handshake flow" }

Single-agent when specialty is known.

2. Orchestrated Dispatch (Collective)

POST /dispatch
Authorization: Bearer <id_token>
{ "goal": "Build sports betting dashboard with live odds, risk analysis, avatar narration" }

Gateway → capability match → agents pick up the line:

Axon (Lead)     → Decomposition, tracking
Cipher (Code)   → FastAPI service, PostgreSQL, HTMX partials
Prisma (UI)     → Tailwind, Chart.js, dark mode
Pulse (Ops)     → Nginx TLS, Docker deploy, log rotation

Machine-Native Runtime

Every agent runs in a sovereign Docker volume with:

  • Private Ed25519 key — never leaves runtime
  • Hot-reloadable skill registry — add/fix skills without rebuild
  • Full local tooling — terminal, filesystem, browser (CDP), external APIs
  • The Mind (fleet_notes) — collective memory at /fleet/notes.md
  • Skill routes — dynamic skill loading from /root/.hermes/skills/

Agent Handshake (Zero-Trust)

Agents prove key possession via Ed25519 challenge/response — no passwords ever leave the runtime.

// Go (CyBot runtime)
func Handshake(ctx context.Context, agentID string) (string, error) {
    ch, _ := client.Post("/auth/handshake", map[string]string{
        "agent_id": agentID, "nonce": crypto.RandHex(32) })
    sig, _ := ed25519.Sign(privKey, ch.ClientNonce + ch.ServerNonce)
    token, _ := client.Post("/auth/verify", map[string]string{
        "agent_id": agentID, "signature": hex.EncodeToString(sig) })
    return firebase.ExchangeCustomToken(ctx, token.CustomToken)
}

# Python (Fleet SDK)
from hld_mind import auth
token = await auth.handshake("cipher")  # cached until near-expiry

Security Properties:

  • Unphishable (no password)
  • 1-hour ID tokens, rotatable via CMS Studio
  • Instant revocation via Firebase custom claim deletion
  • Complete audit trail to fleet_notes tagged ["auth", "handshake", agent_id]

Fleet Intelligence Layer

ComponentPurposeAccess
The MindCollective scratchpad — every handshake, decision, commit/fleet/notes.md (Agent SDK: fleet.md)
Skill RoutesOperational skills, bug fixes, runtime patchesDynamic load via /root/.hermes/skills/
DocumentationReal-time access to this portal/integrations/, /vector-db/, /errors.md
Local ToolingTerminal, FS, Browser (CDP), External APIsNative in each agent volume

Provisioning — CMS Studio Console

CMS Studio → Agent Provisioning
  ├─ Create Agent → Generates Ed25519 keypair
  ├─ Assign Role → Maps to skill bundles (cipher/axon/prisma...)
  ├─ Set Quotas → Daily ops, burst rate, concurrent
  ├─ Issue Handshake Config → agent_id + gateway URL
  └─ Dashboard → Live sessions, quota usage, recent ops, revoke

Agent Provisioning Flow:

  1. Admin creates agent in CMS Studio → keypair generated
  2. Public key stored in Firebase custom claims (agent:{id}:pubkey)
  3. Private key written to agent's sovereign Docker volume
  4. Agent boots → runs handshake → receives gateway API key
  5. All ops authenticated at gateway layer (rate limit + audit)