- Add manifesto with core principles and six portal vision - Add vision & mission with 5-year roadmap and success metrics - Add architecture overview with system diagrams - Add Cece agent mode system prompt for governance layer - Add governance roadmap with KV schema and sprint plan - Add master implementation roadmap tracking 50 tasks 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
16 KiB
16 KiB
BlackRoad OS — Architecture Overview
Version: 1.0 Last Updated: 2025-11-30 Status: Active
System Overview
BlackRoad OS is a distributed AI orchestration platform built on three core layers:
┌─────────────────────────────────────────────────────────────────┐
│ PORTAL LAYER │
│ Lucidia │ RoadWork │ RoadView │ RoadGlitch │ RoadWorld │ BackRoad │
├─────────────────────────────────────────────────────────────────┤
│ GOVERNANCE LAYER │
│ Policies │ Ledger │ Intents │ Claims │ Delegations │
├─────────────────────────────────────────────────────────────────┤
│ INTEGRATION LAYER │
│ MCP │ Tools │ Connectors │ APIs │
├─────────────────────────────────────────────────────────────────┤
│ INFRASTRUCTURE LAYER │
│ Cloudflare │ Railway │ Self-hosted LLMs │ Edge │
└─────────────────────────────────────────────────────────────────┘
Core Architecture Principles
1. Edge-First Design
All latency-sensitive operations happen at the edge via Cloudflare Workers. This includes:
- Request routing
- Authentication/authorization
- Policy evaluation
- Caching
2. Event-Driven Orchestration
The system is built around events, not request/response cycles:
- Every significant action produces a ledger event
- Agents react to events, not direct calls
- Async by default, sync when necessary
3. Governance as Infrastructure
Governance isn't a feature — it's the foundation:
- Every tool call passes through policy evaluation
- Every action is logged to an immutable ledger
- Every permission is explicit and auditable
4. Composable Agents
Agents are small, focused, and composable:
- Single responsibility per agent
- Standard communication protocol
- Orchestrator (Cece) coordinates complex workflows
Layer Details
Infrastructure Layer
The foundation that runs everything.
| Component | Technology | Purpose |
|---|---|---|
| Edge Compute | Cloudflare Workers | Request routing, auth, caching |
| Key-Value Store | Cloudflare KV | Governance objects, config |
| SQL Database | Cloudflare D1 / Railway Postgres | Structured data, analytics |
| Object Storage | Cloudflare R2 | Files, media, backups |
| Backend Services | Railway | APIs, schedulers, workers |
| LLM Inference | Self-hosted Mistral + Claude API | AI operations |
| Message Queue | Railway Redis | Job queues, pub/sub |
Deployment Topology
┌──────────────┐
│ Internet │
└──────┬───────┘
│
┌──────▼───────┐
│ Cloudflare │
│ (Edge) │
├──────────────┤
│ Workers (KV) │
│ Pages (UI) │
│ R2 (Storage) │
└──────┬───────┘
│
┌────────────┼────────────┐
│ │ │
┌──────▼──────┐ ┌───▼────┐ ┌─────▼─────┐
│ Railway │ │ LLM │ │ External │
│ (Backend) │ │ Cluster│ │ APIs │
├─────────────┤ └────────┘ └───────────┘
│ API Gateway │
│ Operator │
│ Agents │
│ Postgres │
│ Redis │
└─────────────┘
Integration Layer
Connects BlackRoad to external systems via Model Context Protocol (MCP).
MCP Architecture
┌─────────────────────────────────────────────────────────┐
│ MCP HOST (Cece) │
├─────────────────────────────────────────────────────────┤
│ ┌─────────┐ ┌─────────┐ ┌─────────┐ ┌─────────┐ │
│ │ Gmail │ │ Drive │ │ Notion │ │ Stripe │ │
│ │ Server │ │ Server │ │ Server │ │ Server │ │
│ └────┬────┘ └────┬────┘ └────┬────┘ └────┬────┘ │
│ │ │ │ │ │
│ ┌────▼────────────▼────────────▼────────────▼────┐ │
│ │ Tool Registry │ │
│ └─────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────┘
Supported Integrations
| Category | Tools |
|---|---|
| Communication | Gmail, Slack, Discord |
| Documents | Google Drive, Notion, Dropbox |
| Development | GitHub, Linear, Jira |
| Finance | Stripe, QuickBooks |
| Infrastructure | Cloudflare, Railway, Vercel |
| Data | Postgres, Redis, Pinecone |
Governance Layer
The brain that ensures safe, auditable operations.
Core Objects
┌─────────────────────────────────────────────────────────┐
│ GOVERNANCE MODEL │
├─────────────────────────────────────────────────────────┤
│ │
│ INTENT ──────► AGENT ──────► TOOL CALL │
│ │ │ │ │
│ │ │ ▼ │
│ │ │ POLICY CHECK │
│ │ │ │ │
│ │ │ ▼ │
│ └─────────────┴────────► LEDGER EVENT │
│ │
└─────────────────────────────────────────────────────────┘
Object Definitions
| Object | Purpose | Key Pattern |
|---|---|---|
| POLICY | Rules that allow/deny/modify actions | policy:{scope}:{id} |
| LEDGER | Immutable audit trail | ledger:{intent_id}:{event_id} |
| AGENT | Registry of workers | agent:{agent_id} |
| INTENT | Tasks and workflow state | intent:{intent_id} |
| CLAIM | Identity assertions | claim:{subject}:{id} |
| DELEGATION | Scoped permissions | delegation:{delegator}:{id} |
Policy Evaluation Flow
┌──────────────┐ ┌──────────────┐ ┌──────────────┐
│ Request │────►│ Policy │────►│ Decision │
│ Context │ │ Engine │ │ Result │
└──────────────┘ └──────────────┘ └──────────────┘
│
▼
┌──────────────┐
│ Actions: │
│ - allow │
│ - deny │
│ - require_ │
│ approval │
│ - transform│
└──────────────┘
Portal Layer
User-facing applications built on the platform.
| Portal | Domain | Key Features |
|---|---|---|
| Lucidia | Personal AI | Persistent memory, preference learning, multi-modal |
| RoadWork | Education | Adaptive curriculum, progress tracking, certification |
| RoadView | Media | Video/image generation, editing, publishing |
| RoadGlitch | Gaming | Procedural content, AI NPCs, adaptive difficulty |
| RoadWorld | Navigation | Context-aware routing, AR overlays, local knowledge |
| BackRoad | Privacy | Encryption, anonymization, secure communication |
Agent Architecture
Agent Hierarchy
┌─────────────────────────────────────────────────────────┐
│ CECE (Governor) │
│ Lucidia-class Orchestrator │
├─────────────────────────────────────────────────────────┤
│ ┌─────────┐ ┌─────────┐ ┌─────────┐ ┌─────────┐ │
│ │ Planner │ │ Builder │ │ Scribe │ │Guardian │ │
│ │ Agent │ │ Agent │ │ Agent │ │ Agent │ │
│ └─────────┘ └─────────┘ └─────────┘ └─────────┘ │
├─────────────────────────────────────────────────────────┤
│ Worker Agents │
│ ┌─────────┐ ┌─────────┐ ┌─────────┐ ┌─────────┐ │
│ │ Email │ │ Code │ │ Data │ │ Media │ │
│ │ Handler │ │ Writer │ │ Analyst │ │ Creator │ │
│ └─────────┘ └─────────┘ └─────────┘ └─────────┘ │
└─────────────────────────────────────────────────────────┘
Agent Classes
| Class | Role | Examples |
|---|---|---|
| Lucidia | Governance & orchestration | Cece |
| Worker | Specialized tasks | EmailHandler, CodeWriter |
| System | Infrastructure operations | Scheduler, Monitor |
| Integration | External tool bridges | GmailServer, StripeServer |
Agent Communication
Agents communicate via structured messages:
{
"message_id": "msg-20251130-abc123",
"from": "cece.governor.v1",
"to": "email.handler.v1",
"intent_id": "int-20251130-xyz789",
"action": "draft_email",
"payload": {
"recipient": "investor@example.com",
"subject": "Follow-up on our meeting",
"context": "..."
},
"constraints": {
"require_approval": true,
"max_tokens": 500
}
}
Data Flow
Request Lifecycle
1. User Request
│
▼
2. Edge Router (Cloudflare Worker)
│
├──► Auth Check (Claims)
│
▼
3. Intent Creation
│
▼
4. Cece Planning
│
├──► Tool Selection
├──► Policy Pre-check
│
▼
5. Agent Execution
│
├──► Tool Call
├──► Policy Evaluation
├──► Ledger Write
│
▼
6. Result Aggregation
│
▼
7. Response to User
Event Flow
┌─────────┐ ┌─────────┐ ┌─────────┐ ┌─────────┐
│ Tool │───►│ Policy │───►│ Ledger │───►│ Index │
│ Call │ │ Check │ │ Write │ │ Update │
└─────────┘ └─────────┘ └─────────┘ └─────────┘
│
▼
┌─────────┐
│ Webhook │
│ Notify │
└─────────┘
Security Model
Authentication
- Users: OAuth 2.0 / OpenID Connect
- Agents: API keys with scoped permissions
- Services: mTLS between internal services
Authorization
┌─────────────────────────────────────────────────────────┐
│ AUTHORIZATION STACK │
├─────────────────────────────────────────────────────────┤
│ 1. Claims - Who are you? │
│ 2. Delegations - What can you do? │
│ 3. Policies - What's allowed in this context? │
│ 4. Audit - What did you do? │
└─────────────────────────────────────────────────────────┘
Data Classification
| Level | Description | Examples |
|---|---|---|
| Public | No restrictions | Marketing content |
| Internal | Org-only access | Design docs |
| Confidential | Role-based access | User data |
| Restricted | Named individual access | Secrets, PII |
Scalability
Horizontal Scaling
| Component | Scaling Strategy |
|---|---|
| Edge Workers | Auto-scaled by Cloudflare |
| API Gateway | Railway auto-scaling |
| Agent Workers | Queue-based scaling |
| Database | Read replicas + connection pooling |
Performance Targets
| Metric | Target |
|---|---|
| Edge latency (P99) | < 50ms |
| API latency (P99) | < 200ms |
| Agent task completion | < 30s |
| Ledger write | < 100ms |
Repository Structure
| Repo | Purpose |
|---|---|
blackroad-os-core |
Desktop UI, auth, identity |
blackroad-os-api |
API gateway, REST endpoints |
blackroad-os-operator |
Job scheduler, background workers |
blackroad-os-agents |
Agent implementations |
blackroad-os-infra |
IaC, deployment configs |
blackroad-os-docs |
Documentation (this repo) |
blackroad-os-web |
Web frontend |
blackroad-os-prism-console |
Admin dashboard |
lucidia-core |
AI reasoning engines |