R
Role-aware orchestration
Route tasks to the correct agent persona based on skill, channel, permission, and business context.
AI workforce runtime
The agent brain that knows who should do what.
A Mike Merlino version of the enterprise agent runtime idea: role-aware AI teammates, memory, permissions, audit trails, and deployment across Telegram, Slack, web, SMS, GHL, and internal tools.
mission-control/runtime
Oliverorchestration + dependency routingactive
Carlosimplementation + QA executionqueued
Sherlockresearch, evidence, competitive intelreading
HermanVPS3 outbound + reporting operatorlive
/route client=DWS task="enterprise visibility sprint" policy=RBAC memory=project scoped audit=on
22+agent roles
6channels
∞memory graph
Runtime architecture
Five layers. One decision engine.
Modeled after the agentbrain.sh category, but written for Mike’s ecosystem: Hermes, Herman, OpenClaw-style rosters, memory providers, outreach engines, and client deliverables.
M
Memory spine
Profile recall, project containers, session ingest, durable notes, and procedural skills working together.
A
Audit-first execution
Every command, scrape, deploy, digest, and outreach artifact can be traced back to source data and operator intent.
C
Channel deployment
Telegram, Discord, Slack, SMS, web dashboards, client pages, and internal APIs from one operator layer.
Security model
Built for operators who cannot afford chaos.
Secrets stay out of the chat. Skills define procedure. Memory is scoped. Risky installs are reviewed. Outputs are path-grounded and verifiable.
PolicyRBAC / agent scopedRequired
Memoryprofile + project tagsHybrid
DeployVercel preview + artifact pathLogged
Secretsenv file / vault referenceHidden
What it replaces
Not another prompt box.
This is the layer between messy business operations and actual AI execution: who can act, what they know, where they can post, what they are allowed to touch, and what proof they return.
RBACAudit LogsAgent TeamsMemory ContainersHuman Approval Gates
Agent roster
Installed teammates, not generic bots.
Each role has a job, memory scope, and execution lane. This is where Mike’s “agent workforce” becomes controllable infrastructure.
Oliver
Breaks down complex jobs, assigns specialists, checks dependencies, and returns the executive synthesis.
Carlos
Turns specs into working pages, automations, artifacts, and QA-ready deliverables.
Gino
GHL, CRM workflows, follow-up automation, pipeline logic, and client ops plumbing.
Sherlock / Raven
Competitive intelligence, lead due diligence, citations, proof gathering, and audit reports.
Spielberg
Ranking Reels scripts, video SEO, local proof clips, shorts, captions, and sales assets.
Herman
VPS3 operator for lead discovery, enrichment, scoring, Vercel previews, and concise reporting.
Deployment flow
From business intent to shipped work.
The runtime is valuable because it converts rough voice/text commands into safe, useful, reviewed execution.
01 IntakeTelegram, voice, Slack, client portal, form, webhook.
02 ClassifyIntent, risk, client, project, agent lane, memory scope.
03 RouteAssign the agent/team with the right tools and constraints.
04 ExecuteBuild, scrape, enrich, deploy, draft, analyze, or schedule.
05 ProveReturn links, screenshots, counts, paths, logs, next actions.