Harbor
Skip to content
System Architecture

Five layers.
One governed surface.

Harbor is structured like an operating system. Applications sit on top, the decision chain runs in the middle, governance wraps action, and the Nexus language underneath gives every agent the same shared semantics.

Architecture status
Live operating model

This page is the buyer-facing read of how Harbor is structured, not just what each agent does in isolation.

Active agents
3
1 planned
Latest chain scan
Waiting
Current equity
Live paper-trading runtime
Public language
Fabric
Ontology internally
Immersive Graph

The governed chain, as a live orbit.

This is the circular system view of Harbor’s decision chain: Macro feeding regime, Sentiment feeding flow, Signal scoring conviction, Valuation anchoring price, Risk gating capital, and Execution closing the loop. It’s interactive, it’s immersive, and it belongs on Architecture.

Chain telemetry
Macro Agent
1 / 6
Upstream
1
Outputs
2
Downstream
4
Live Scan
Listening
Active
Macro
Sentiment Agent
Signal Agent
Risk Agent
Active agent
Macro Agent
Step 1

Regime detection and structural break context.

Receives from
System Context
Writes into Nexus
RegimeAdvisory
Runtime posture
Observe First Implemented

Implemented in Harbor, but not visible in the latest stored chain scan.

Downstream influence
Sentiment AgentLinked
Signal AgentLinked
Risk AgentLinked
Execution AgentLinked
State 1
Pass
State 2
Block
State 3
Review
Platform Stack

From surface to substrate.

The stack reads top-down because that is how a buyer encounters Harbor first, but it reasons bottom-up in practice: foundation, governance, language, agents, then surfaces.

01
Applications
User-facing surfaces
Portfolio intelligenceTrade reconciliationSignal lineageRisk governanceDecision auditFabric explorer
6 surfaces
02
Agent Decision Chain
Observe → Score → Govern → Act
MacroSentimentSignalValuationRiskExecution
6 agents
03
Nexus Language + Toolchain
Domain primitives
SignalAssetSleeveConstraintPositionTradeRiskRegime
8 types
04
Risk & Governance
Policy enforcement layer
Governed sizingExposure limitsDrawdown controlsKill switchesAudit visibility
Enforcing
05
Enterprise Data + Logic + Action
Foundation layer
harbor_shadow-4.pyharbor_paper.dbIB Gateway
Foundation
Surface
Applications make Harbor legible.

Dashboards, reports, and system views are where the chain becomes usable. They translate governed outputs into something an operator or buyer can actually read.

Chain
The decision fabric narrows commitment.

Each agent adds one kind of judgment, then passes a tighter object forward. Observation becomes scoring, scoring becomes valuation, and valuation only becomes action after governance.

Language
Nexus keeps meaning stable.

Shared primitives stop every page and every agent from inventing a slightly different definition of signal, risk, position, or trade. That consistency is part of the product.

Decision Flow

From signal to settlement.

This is the system-wide view: one shared context bus above the chain, the six agents in sequence, and the persistence, governance, and narrative surfaces underneath.

Shared semantic bus
DecisionContext

MacroCtx · SentimentCtx · SignalCtx · ValuationCtx · RiskCtx · ExecutionCtx

Agent memory

agent_memory · scan_digest · sleeve_ledger

Governance

PASS · BLOCK · REVIEW · approval queue

Claude narrative

Natural-language inspection over the same chain

Infrastructure

How data moves through the system.

The architecture only works because the same few infrastructure patterns repeat everywhere: shared context, persistent memory, governance, broker connectivity, operator surfaces, and human-readable narrative.

Context Bus
DecisionContext

Shared context is the central contract. Agents do not whisper to each other directly; they publish structured context into a common surface.

Memory
Agent memory

Harbor persists state across scans with `agent_memory`, `scan_digest`, and `sleeve_ledger`, so later decisions have continuity instead of amnesia.

Governance
3-state engine

Every proposal ends in PASS, BLOCK, or REVIEW. Governance is not a decorative wrapper around the chain; it is the operating boundary around capital.

Broker
IBKR bridge

Execution reaches the market through one governed path. Routing, fills, and reconciliation all happen through the same broker bridge.

Surfaces
Operator applications

Dashboard, signal lineage, audit surfaces, and the system canvas are all applications sitting above the same chain and the same language.

Narrative
Claude interface

Natural-language access belongs in the architecture view because it is how humans inspect reasoning, not a separate intelligence stack bolted on later.

System Graph

The live visual anchor belongs here too.

You were right to call this out. Architecture should still carry a live graph view of the system so the page feels like the operating model, not just a text explanation. Nexus remains the deeper semantic breakdown, but the graph belongs on this page as a flagship system view.

Live Harbor graph

System nodes, cluster groupings, and cross-layer connections rendered from Harbor's live canvas payload.

Loading nodesLoading connectionsLoading zones
Live topology
Preparing the system graph.
Platform Extraction

Domain-agnostic by design.

One of the clearest buyer stories in Harbor is that the infrastructure is portable. The finance-specific logic sits on top of a much more general governed decision framework.

Layer 1 · Harbor Core
Domain-agnostic infrastructure

Multi-agent orchestration, shared context, governance, memory, approval queue, operator surfaces, and narrative inspection are not finance-specific ideas. They are the reusable core.

DecisionContextAgent interface3-state governanceMemory systemApproval queueNarrative layer
Layer 2 · Harbor Finance
Finance-specific implementation

The chain, broker path, sleeve logic, Kelly sizing, and paper-trading runtime are the finance implementation. That is where Harbor becomes an equity operations system.

IBKR bridgeSleeve ledgerKelly sizingMonte CarloSignal engineRisk policy
Nexus Handoff

Shared language deserves its own surface.

Architecture explains the operating model. Nexus explains the nouns, links, and contracts that keep that operating model coherent. You should feel the distinction when you move between the two pages.

Explore the Chain

See every agent up close.

Walk the decision chain from Macro to Execution, then inspect the shared language underneath it on the Nexus surface.