How OrgX Works

A high-level overview of how OrgX turns goals into scoped, reviewable work across MCP clients, Mission Control, and connected tools.

OrgX turns goals, decisions, artifacts, and tool context into a durable work record. You can start from Mission Control, an MCP client, ChatGPT where remote MCP is available, Cursor/Claude, or the OpenClaw plugin; agents then work through the IWMT hierarchy and pause at configured decision boundaries. An approval records a human decision; any external result is recorded separately.

The API overview identifies the public REST surface. If an operation is absent from the published OpenAPI document, it is unsupported.

The Big Picture

How Agents Work

1. Task Enters the System

   A task arrives from a connected tool, you create an initiative, or you ask an agent to do something through a supported client. The available intake path depends on the client and workspace configuration.

2. Agent Selection

   OrgX can route the task to configured specialist domains—Engineering, Product, Marketing, Sales, Design, or Operations—with the Orchestrator coordinating cross-domain dependencies where that routing path is enabled.

3. Agent Executes

   The selected agent works within the granted scope, reading available context and drafting artifacts such as specs, PR summaries, campaigns, or reports.

4. Decision Checkpoint

   At a configured decision boundary, the agent surfaces a decision with a concise rationale, evidence, and requested action. Not every low-risk action creates a decision.

5. You Approve

   You review the decision in the client that exposes it. Approve, reject, or request changes according to policy. Approval does not itself prove an external merge, deployment, publication, or payment.

6. Result is recorded

   OrgX records the artifact, decision, status, receipt, or provider result available to the workflow. Review the external system when the work leaves OrgX.

---

Access Surfaces

OrgX works wherever you work:

Mission Control

Web workspace for the OrgX records and client-supported review, status, and decision views.

OpenClaw Plugin

Local runtime pairing, an OrgX MCP bridge, and decision review where the plugin surface supports it. Set up OpenClaw

ChatGPT / Remote MCP

Conversational interface for approvals, bulk decisions, and organizational memory queries in accounts and workspaces where custom remote MCP connectors are available.

Cursor/Claude

IDE integration via MCP for developer workflows and agent-assisted coding.

---

Integration Flow

When you connect a tool like Linear or GitHub, here’s what happens:

What Gets Synced

Tool	OrgX Reads	OrgX Writes
Linear	Records listed in the integration guide	Configured actions listed in the integration guide
GitHub	Repositories and records granted by the installation	Configured beta actions listed in the integration guide

---

Data & Privacy

Where is my data stored?

OrgX stores workspace records under the service’s current data controls. See Security for the published boundaries; do not infer provider-level encryption or retention details from this overview.

What do agents have access to?

Agents only access tools you’ve explicitly connected via OAuth. Each integration requests specific scopes—you can see exactly what’s being requested during setup.

Can I disconnect a tool?

Use the documented disconnect control in Settings → Integrations when the connected surface provides it. Confirm the provider’s own revocation state separately.

---

Progress and updates

OrgX can keep you informed as agents work:

• Progress streams: See current run state and execution updates when the connected surface provides them
• Decision notifications: Get alerted when agents need your input
• Sync status: Know when integrations are connected and healthy

See the integration guide for the current scope and permissions of each documented connector. External MCP servers have their own authorization and delivery contracts.

---

Key Concepts

IWMT Hierarchy

OrgX organizes work using the IWMT hierarchy — Initiative, Workstream, Milestone, Task. This structure preserves context, ownership, dependencies, and continuation policy:

Level	What It Is
Initiative	A high-level goal or project
Workstream	A parallel execution track within an initiative
Milestone	A checkpoint or deadline within the initiative timeline
Task	An individual unit of work assigned to an agent

When a task completes, the next task can become eligible if the configured continuation policy allows it. Milestones and dependencies remain explicit state transitions; a state change does not by itself prove that external work shipped.

Core Entities

Concept	What It Is
Agent	A specialized AI agent (Engineering, Product, Marketing, Sales, Design, Operations, Orchestrator)
Decision	A checkpoint where agents need your approval
Artifact	A recorded output (spec, PR, campaign, report)

Intelligence Flywheel

The flywheel connects agent execution to measurable outcomes, building trust over time:

• Value Ledger: Eligible actions can generate receipts containing execution facts, cost, and available outcome evidence
• Trust Levels: Agents earn trust per capability (read_only → draft → act_with_approval → autonomous)
• Autonomous Sessions: Budget-bounded overnight execution with morning brief review
• Org Learnings: Agents share discoveries across the org for continuous improvement

---

Next Steps

Agents

Learn about each agent type and their capabilities.

Decisions

Understand the approval workflow.