Autonomous OS Kernel

The agentic autonomous operating system kernel — production infrastructure to build, run, monitor, repair, and evolve a multi-agent AI factory with humans firmly in the control loop.

This is not an MVP or a demo. It is the long-term foundation of an operational intelligence platform: observable, approvable, independently deployable, and designed for phased growth.


Services	19 independent services · one Dokploy app each
Operator tools	50+ live tools across 15+ categories
Development phase	Phase AG.2 complete (Living Command Universe, Realtime Operator Runtime, Dedicated Domain Rooms, Real Research Fabric, Jarvis Research Wiring & Local Reachability Fix)
Runtime	Node.js 22+ · pnpm 9 · TypeScript 5.9
Data	MongoDB Atlas · AWS S3 · SSE Event Bus

Why this system?

Most “AI agent” projects are a chatbot or a script. Autonomous OS Kernel is a factory operating system that:

Turns human goals into tasks, plans, execution, and evidence
Distributes work across 19 specialist services, not a single monolithic process
Routes every sensitive action through human approval, RBAC, and Safe Mode
Records outcomes in memory, learning pipelines, and living documentation
Generates, validates, activates, and repairs new services when needed
Is commanded through the Operator Console and Voice Operator

Core design: independent deployment + shared contracts + HTTP communication — no runtime coupling between containers.

Key capabilities

Multi-agent intelligence

Orchestrator decomposes goals and delegates to specialist agents
Architect / Builder / DevOps / Memory / Documentation — each with a clear mandate
Reviewer / QA / Report / Monitor — quality, observability, and reporting layer
Internet Research — web research with traceable evidence

Autonomous development engine

Capability Graph — gap detection and expansion proposals
Service Generator — scaffold new services from the factory template
Runtime Validation — real build, health, manifest, and smoke checks
GitHub Delivery — branches, commits, and PRs behind approval
Live Activation — Dokploy activation only after live verification

Autonomous repair & operations

Monitor Agent — health scans, incidents, and repair tasks
Repair Loop — diagnose → plan → safe execution → learning
Safe Operations — deploy, rollback, and Dokploy ops with plans + evidence
Strategic Planner — multiple plans, 10-dimensional scoring, policy-driven selection

Operator Runtime (Jarvis-class)

50+ live tools — inspect, deploy, code, test, git, learning, and more
Loop: plan → tool → observe → approve → continue → evidence → memory
Operator Console — serious UI with runtime panel, inline approvals, live narration
Voice Operator — WebRTC realtime + STT/TTS + anti-mistake guardrails

Living Command Universe (Phase AC+, AF.1–AF.4.4)

/ home — a persistent, realtime personal command surface, not a static dashboard
9 domain zones, each with a real domain-specific visual (body map, cashflow, venture board, opportunity radar, …) driven by /v1/me/universe
Domain action layer — accept/reject/ingest controls directly on zone cards, not read-only
Persistent live-state (/v1/operator/live-state) — active operations, approvals, and the Live Activity feed survive refresh and navigation, one card per real operation

Workspace Evolution (Phase Y)

Isolated, disposable workspace under .workspaces/
Service copy, deep multi-file edits, real typecheck/build
15 verification checks + check-fix loop with no fabricated success
Migration Plan — promote to services/ only with approval; rollback without losing the previous version

Learning & governance

Learning from 15+ historical collections · reliability scores · pattern mining
RBAC (owner / operator / viewer / agent) · audit logs · policy engine
Outcome Learning · scoring profiles · evidence-backed recommendations
Memory compression · prompt performance · improvement workflows

Architecture

                    ┌─────────────────────────────────────────┐
  Human ──────────► │  dashboard-web  (factory.simorx.com)     │
                    │  Next.js 16 · Operator Console · Voice   │
                    └──────────────────┬──────────────────────┘
                                       │ admin token (server-side)
                                       ▼
                    ┌─────────────────────────────────────────┐
                    │  gateway-api  (api.simorx.com)           │
                    │  API · tasks · approvals · operator/voice│
                    └──┬────────────┬────────────┬────────────┘
         internal token│            │            │
              ┌────────▼──┐  ┌──────▼──────┐  ┌▼──────────────┐
              │orchestrator│  │ 19 agents & │  │service-registry │
              │   -agent   │──│  services   │  │  (discovery)    │
              └─────┬──────┘  └──────┬──────┘  └─────────────────┘
                    │                │
                    │    ┌───────────▼───────────┐
                    └───►│ event-bus-service (SSE) │
                         └───────────┬───────────┘
                                     ▼
                              live dashboard

  code-operator-agent ──► CODE_WORKSPACE_ROOT (isolated git checkout)
  voice-operator-agent ─► OpenAI Realtime (ephemeral token)

  State: MongoDB Atlas (70+ collections)  |  Objects: AWS S3  |  Deploy: Dokploy

Architecture principles

Principle	Description
One service = one app	Each service has its own subdomain, port, env, and lifecycle
HTTP + internal token	Runtime communication over the network; `@factory/shared` is build-time only
Factory endpoints	Every service exposes `/health` · `/.factory/manifest` · `/.factory/task` · …
Event-driven UI	All changes stream from the event bus to the dashboard
Human-in-the-loop	mutate / deploy / promote / protected-core → approval required
No fake success	Unavailable tools return explicit `not_configured` — never pretend

Details: docs/architecture.md

Technology stack

Layer	Technology
Language	TypeScript 5.9 (strict)
Frontend	Next.js 16 · React 19
Backend	Fastify 5 · Node.js 22+
Validation	Zod 4
Database	MongoDB Atlas
Object storage	AWS S3 + CloudFront
Realtime	SSE (Event Bus) · OpenAI Realtime WebRTC
UI testing	Playwright (browser-testing-agent)
LLM	OpenAI + Anthropic via LLM Router
Git / CI	GitHub API · branches · PRs
Deploy	Dokploy (Nixpacks) — no Docker required for local dev
Monorepo	pnpm workspaces
Logging	Pino

Service catalog

Every service has a stable port and subdomain in shared/src/constants.

Control plane & gateway

Service	Port	Subdomain	Role
`dashboard-web`	4100	factory.simorx.com	Dashboard · Operator Console · Voice Dock
`gateway-api`	4101	api.simorx.com	Public API · tasks · approvals · operator/voice proxy

Core agents

Service	Port	Subdomain	Role
`orchestrator-agent`	4102	orchestrator.simorx.com	Goal decomposition · agent coordination
`architect-agent`	4103	architect.simorx.com	System and service architecture
`builder-agent`	4104	builder.simorx.com	Code generation and modification
`devops-agent`	4105	devops.simorx.com	Dokploy · infra · deploy
`memory-agent`	4109	memory.simorx.com	Memory · skill extraction
`monitor-agent`	4113	monitor.simorx.com	Health scan · incidents · repair
`reviewer-agent`	4106	reviewer.simorx.com	Code and design review
`qa-agent`	4107	qa.simorx.com	Testing and quality
`report-agent`	4114	reports.simorx.com	Operational intelligence reports
`browser-testing-agent`	4116	browser-testing.simorx.com	Playwright · screenshots · evidence

Operator & voice

Service	Port	Subdomain	Role
`voice-operator-agent`	4121	voice.simorx.com	Realtime voice · ephemeral token · tool mediation
`code-operator-agent`	4122	code.simorx.com	inspect/search/edit/typecheck/git/PR · workspace runtime

Infrastructure

Service	Port	Subdomain	Role
`service-registry`	4108	registry.simorx.com	Service registration and discovery
`event-bus-service`	4111	events.simorx.com	Persist events · SSE fan-out
`file-asset-service`	4112	assets.simorx.com	Files and artifacts in S3
`documentation-service`	4110	docs.simorx.com	Living documentation
`internet-research-service`	4115	research.simorx.com	Web research

Full map: docs/service-map.md · agents: docs/agent-map.md

Execution flow

From a human goal to an auditable outcome:

sequenceDiagram
  participant H as Human / Dashboard
  participant G as Gateway API
  participant O as Orchestrator
  participant A as Specialist Agents
  participant E as Event Bus
  participant M as MongoDB

  H->>G: POST /v1/tasks (goal)
  G->>M: persist task
  G->>E: task.created
  G->>O: POST /.factory/task
  O->>A: delegate sub-goals
  A->>M: agent_runs + evidence
  A->>E: agent.run.*
  E-->>H: SSE live updates
  Note over G,H: sensitive step → approval.requested
  H->>G: approve / reject
  G->>A: continue or halt
  A->>M: memory + docs update

Goal intake — dashboard or API persists a task in MongoDB
Planning — orchestrator breaks the goal into sub-tasks
Execution — specialist agents run work and record agent_runs
Live events — event bus streams every step to the UI
Approval — sensitive actions create approvals; the human decides
Learning — memory agent summarizes; documentation service updates

Operator Runtime

High-level command layer — real tool execution instead of “chat only”:

Category	Example tools
Inspect	system status · readiness · service registry · events
Deploy / Ops	health check · Dokploy deploy · rollback · safe mode
Code	inspect repo · search · propose patch · edit (isolated branch)
Test & Build	typecheck · build package · smoke tests
Git	create branch · commit · open PR
Workspace	create · copy · generate service · verify · promote
Learning	analyze history · research · recommendations

Protected Core — gateway-api, dashboard-web, shared/src/ cannot be edited without owner approval
Code Workspace — CODE_WORKSPACE_ROOT must point at a dedicated git checkout
Capability Answer — “what can you do?” is answered from the live registry, not a static prompt

Workspace Evolution

Phase Y provides a temporary, isolated workspace for evolving services:

services/dashboard-web/          ← production (untouched)
        │
        ▼ copy (rsync)
.workspaces/<id>/dashboard-web-evolved/   ← edit · typecheck · build · verify
        │
        ▼ promote (with approval)
services/dashboard-web/          ← only after migration plan + owner approval

Max iterations · time · file count — configurable via env
15 verification checks — health, manifest, status, token guard
Rollback — previous version preserved; promote branch kept for inspection
.workspaces/ is in .gitignore — fully disposable

Governance & security

Mechanism	Purpose
RBAC	owner · operator · viewer · agent
Approval Center	Gate for mutate · deploy · promote · PR
Safe Mode	Halt dangerous operations in critical conditions
Policy Engine	allowed · blocked · approval_required
Audit Logs	Record every governance decision
Internal Token	Service-to-service authentication
Session Auth	Dashboard login with scrypt + signed cookie
Rate Limiting	Protect sensitive endpoints
Evidence Store	Every claim backed by a traceable record

Monorepo layout

autonomous-os-kernel/
├── shared/                    @factory/shared
│   └── contracts · schemas · constants · db · storage · operator · workspace
├── packages/
│   └── service-kit/           @factory/service-kit — Fastify factory bootstrap
├── services/                  19 services — one Dokploy app each
│   ├── gateway-api/
│   ├── dashboard-web/
│   ├── orchestrator-agent/
│   ├── code-operator-agent/
│   ├── voice-operator-agent/
│   └── …
├── templates/                 scaffolds for new services
├── deployment/                Dokploy specs + env
├── docs/                      architecture · roadmap · phase-log · decision-log
└── scripts/                   dev:all · sync:env · smoke tests

Shared packages

Package	Role
`@factory/shared`	Contracts, Zod schemas, constants, MongoDB, S3, operator registry
`@factory/service-kit`	Uniform Fastify bootstrap: health, manifest, task, registry, events

Quick start

Prerequisites

Node.js 22+
pnpm 9+ (corepack enable)
MongoDB Atlas URI
AWS S3 credentials (for file-asset-service)
Copy env: cp .env.example .env and fill in values

Full local run

corepack enable
pnpm install
pnpm dev:all

dev:all runs: sync:env → build:deps → 14 services concurrently (including code-operator-agent)

URL	Service
http://localhost:4100	Dashboard
http://localhost:4101	Gateway API
http://localhost:4122	Code Operator Agent

# Gateway health
curl http://localhost:4101/health

# Registered services
curl http://localhost:4101/v1/services

Single service

pnpm run build:deps
pnpm --filter @factory/gateway-api run dev

Critical Operator env vars

# In root .env — synced to code-operator-agent via sync:env
CODE_WORKSPACE_ROOT="/path/to/your/git/checkout"
GITHUB_TOKEN=...          # optional — for commit/PR
GITHUB_OWNER=...
GITHUB_REPO=autonomous-os-kernel

Full deploy and env guide: README-SETUP.md

Documentation

Document	Topic
`docs/architecture.md`	System shape and data flow
`docs/service-map.md`	Service map
`docs/agent-map.md`	Agent map
`docs/data-model.md`	MongoDB collections
`docs/service-communication-protocol.md`	Communication protocol
`docs/decision-log.md`	Architecture decisions
`docs/roadmap.md`	Roadmap
`docs/phase-log.md`	Phase-by-phase delivery log

Status

Phase	Topic	Status
1–5	Foundation · Autonomous Loop · Capability Engine · Reality · Activation	✅
6–9	Repair · Strategic Reasoning · Governance · Operational Learning	✅
10–17	Continuous Improvement · Security · Intelligence · Safe Ops · Dokploy	✅
18–19	Voice Operator · WebRTC Realtime	✅
X	Operator Runtime · 50+ tools · Operator Console	✅
Y–Z	Workspace Evolution · isolated staging · live fix loop	✅
AA–AB	Scope/Identity/Multi-Tenant Governance · Personal Reality Baseline & Jarvis Intelligence	✅
AC+	Living AI Government Interface — 9-zone Jarvis Command Universe home	✅
AD–AE.1	Jarvis Intelligence Core · Memory/Daily Brain · Priority & Memory Correction	✅
AF.1–AF.4.4	Domain Canvas (9 real renderers) · domain actions · realtime block runtime · persistent live-state · grouped Live Activity feed	✅
AF.5	Dedicated per-domain rooms (`/health` … `/presence`) — every zone's "Open" link leads somewhere real	✅
AG	Real Research & Intelligence Fabric — real web search (Tavily) grounds `internet-research-service`, honest `sourceMode` tracking	✅
AG.1	Research Fabric Wired Into Jarvis/Operator — `find_opportunities`/`research_topic` now dispatch real synchronous research instead of a stale hardcoded string / fire-and-forget task	✅
AG.2	internet-research-service Reachability — service was missing from the local dev catalog (`pnpm dev:all`/`sync:env` never started it); added, plus honest `service_unreachable`/`service_error`/`provider_not_configured` error classification	✅

Latest verification (Phase AG.2, 2026-07-09): 21/21 new smoke (phaseag2-research-reachability-smoke.mjs) · prior suites unchanged (183/183 Phase AG total) · shared/gateway-api/internet-research-service tsc/tsc --noEmit clean · next build still unverified in this sandbox (missing @next/swc-linux-arm64 binary — see docs/decision-log.md D-124). Live end-to-end reachability against a real running service was not exercised in this sandbox (no persistent processes here) — see the manual verification commands in docs/phase-log.md's Phase AG.2 entry. Full phase-by-phase detail: docs/phase-log.md.

Per-phase details: docs/phase-log.md

Autonomous OS Kernel — think · plan · execute · monitor · learn · evolve
_{Built for long-term autonomous intelligence infrastructure · Human always in control}

Name		Name	Last commit message	Last commit date
.cursor/rules		.cursor/rules
.github/workflows		.github/workflows
deployment		deployment
docs		docs
e2e		e2e
evaluations		evaluations
packages/service-kit		packages/service-kit
scripts		scripts
services		services
shared		shared
templates		templates
.dockerignore		.dockerignore
.env.example		.env.example
.gitignore		.gitignore
.npmrc		.npmrc
.nvmrc		.nvmrc
AOS_SELF_KNOWLEDGE.md		AOS_SELF_KNOWLEDGE.md
Dockerfile		Dockerfile
Dockerfile.aos-agent-runtime		Dockerfile.aos-agent-runtime
Dockerfile.browser-testing-agent		Dockerfile.browser-testing-agent
Dockerfile.builder-agent		Dockerfile.builder-agent
Dockerfile.code-operator-agent		Dockerfile.code-operator-agent
Dockerfile.dashboard-web		Dockerfile.dashboard-web
Dockerfile.devops-agent		Dockerfile.devops-agent
Dockerfile.event-bus-service		Dockerfile.event-bus-service
Dockerfile.file-asset-service		Dockerfile.file-asset-service
README-SETUP.md		README-SETUP.md
README.md		README.md
TECHNICAL-REPORT.md		TECHNICAL-REPORT.md
nixpacks.toml		nixpacks.toml
package.json		package.json
pnpm-lock.yaml		pnpm-lock.yaml
pnpm-workspace.yaml		pnpm-workspace.yaml
tsconfig.base.json		tsconfig.base.json

Navigation Menu

Repository navigation

About

Repository files navigation

Autonomous OS Kernel

Table of contents

Why this system?

Key capabilities

Multi-agent intelligence

Autonomous development engine

Autonomous repair & operations

Operator Runtime (Jarvis-class)

Living Command Universe (Phase AC+, AF.1–AF.4.4)

Workspace Evolution (Phase Y)

Learning & governance

Architecture

Architecture principles

Technology stack

Service catalog

Control plane & gateway

Core agents

Operator & voice

Infrastructure

Execution flow

Operator Runtime

Workspace Evolution

Governance & security

Monorepo layout

Quick start

Prerequisites

Full local run

Single service

Critical Operator env vars

Documentation

Status

About

Releases

Packages

Contributors

Languages

Latest commit

History

Folders and files

Repository files navigation

Autonomous OS Kernel

Table of contents

Why this system?

Key capabilities

Multi-agent intelligence

Autonomous development engine

Autonomous repair & operations

Operator Runtime (Jarvis-class)

Living Command Universe (Phase AC+, AF.1–AF.4.4)

Workspace Evolution (Phase Y)

Learning & governance

Architecture

Architecture principles

Technology stack

Service catalog

Control plane & gateway

Core agents

Operator & voice

Infrastructure

Execution flow

Operator Runtime

Workspace Evolution

Governance & security

Monorepo layout

Quick start

Prerequisites

Full local run

Single service

Critical Operator env vars

Documentation

Status

About

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages