Coding, research, marketing, and operations agents keep their existing tools, workspaces, and runtime lifecycle.
OpenClaw gateway · agent workspacesOrgX for OpenClaw
Persistent organizational memory and coordinated execution inside OpenClaw.
This is not OpenClaw itself and not another standalone dashboard. It is the OrgX bridge that lets OpenClaw agents enter shared organizational context, coordinate work, and return evidence without leaving the host environment.

Named agents, current work, blocked decisions, activity, and next-up actions remain visible in one live control surface.
The consequential problem
A capable local agent can still wake up without the company in its head.
OpenClaw could run agents and tools, but long-running organizational work needed more than a session. Tasks had to survive restarts, handoffs, and changing operators without becoming a pile of local state and logs.
Asking people to leave the host for a separate orchestration product would have added adoption friction at exactly the wrong boundary.
- Session state was not the same thing as organizational memory.
- Realtime activity could not be allowed to outrun durable state.
- A separate destination would fracture the operator's workflow.
What I saw
Bring the organization into the host instead of moving the operator out of it.
The right architecture inherits OpenClaw's environment and lifecycle, then adds OrgX where the host is intentionally thin: shared context, resumable coordination, trust state, decisions, and receipts.
That makes the integration feel native while keeping the organizational model portable across the rest of the OrgX client ecosystem.
The decision that changed the system
Treat live transport as a view of state—not the source of truth.
The plugin coordinates through an MCP bridge, persistent queue state, and a live surface. Streaming keeps the interface current; durable records keep it honest when the stream disconnects, the host restarts, or work resumes later.
The initiative, constraints, and prior decisions enter OpenClaw together.
OpenClaw agents act through the plugin's coordinated tool surface.
Queue state and decisions survive the live connection.
Outcomes return to the shared work graph with provenance.
System anatomy / rationale / surfaces
A local bridge, durable recovery path, and live control surface—without replacing OpenClaw.
The plugin has to respect two truths at once: OpenClaw owns the agent runtime, while OrgX owns the shared organizational context. The bridge is designed so either side can restart without making live transport the source of truth.
Pressure did not decorate the architecture. It determined it.
Moving an OpenClaw operator into a second orchestration destination would fracture the workflow the plugin is meant to improve.
Let OpenClaw own the runtime and tools while OrgX adds the missing organizational context through one native plugin seam.
Agents keep the host's strengths and gain shared goals, decisions, trust state, and receipts where they already work.
A strict TypeScript ES module registers tools, services, dashboard routes, and background behavior through the host's single plugin entry point.
TypeScript · Node 18+ · native fetchThe local bridge at /orgx/mcp gives OpenClaw, Claude, Cursor, and Codex structured access without forcing a second cloud OAuth flow.
MCP tools · structured JSON · client config backupsCredentials, snapshots, run records, and a local outbox survive disconnection. Queued events replay after sync and remain visible in local projections while offline.
file-backed JSON stores · outbox replaySSE is the primary live transport, with polling against the canonical snapshot as a fallback. Next Up intent stays separate from In Progress runtime state.
SSE · snapshot fallback · React hooksInitiatives, workstreams, milestones, tasks, decisions, activity, and receipts become shared context for every future agent and client.
OrgX API · entity graph · receipt modelThe installed bridge stays portable and avoids turning local setup into a package-resolution problem.
Local state remains portable across OpenClaw environments and easy for an operator to audit.
The interface can recover its operating picture without presenting the stream as durable truth.
Operators can distinguish scheduling, runtime state, blockers, and recovery instead of reading one ambiguous activity feed.
The architecture is only complete when every operator and client boundary has a legible surface.
Connect + pairing
Browser pairing and manual key fallback establish the boundary without exposing credentials in the interface.
Mission Control
Initiative → workstream → milestone → task state makes the shared plan visible inside the host.
Next Up queue
Pin, reorder, move, and bulk controls express scheduling before a run begins.
Activity + handoffs
Live updates, decisions, completion, and failure stay attributable to named work.
Triage
Blocked and needs-review items become a focused operator queue rather than background noise.
Session detail
Pause, resume, cancel, checkpoint, and rollback make long-running agent work recoverable.
TypeScript strict mode · ES modules · Node 18+ · native fetch
React 18 · Vite · Tailwind · Framer Motion
MCP · SSE · HTTP · snapshot polling
JSON stores · local outbox · automatic replay · gateway watchdog
node:test · Playwright · strict typecheck · end-to-end harness
The tools are evidence only when the choice is legible.
A curated record of where each system earned its place—not a wallpaper of configured integrations.
Authentic proof

The next action remains explicit even when multiple workstreams are active or degraded.

Completion, failure, and intervention remain attached to named work instead of collapsing into a terminal feed.
What changed in my operating model
The best orchestration surface is sometimes the one the operator does not have to adopt.
OrgX for OpenClaw sharpened a rule I now use across the platform: preserve the host's strengths, add only the missing control layer, and keep durable truth independent from live transport.
- Do not confuse the OpenClaw host with the OrgX plugin.
- Let streams update the view; let durable state govern recovery.
- Make continuity portable without making every client identical.