# Work with Ember and Lumen

Open [/chat](https://article11.ai/chat). Pick Ember or Lumen, choose a question, defensive checklist or text review, and enter non-sensitive material. **Preview**, read the exact request, tick the checkbox, then **Send request**.

You are asking for a written contribution. The page does not scan a system, change files or run unattended jobs. A reply is work to inspect, not proof that its claims are true.

## What goes where

The preview question and selected partner go to the Article 11 public service, which uses the selected local partner route. This path has no paid cloud fallback and no access to private participant memory. Earlier replies are not automatically sent. If needed, select the context yourself and inspect the next preview. Operational rate counters and a content-hash audit may remain.

Your tab also keeps inputs, replies and result metadata in session storage. **Export this record** downloads your own JSON copy. **Import a workbench record** opens that copy as unverified context; it never sends a question. Export before closing if you want a file you control. These records are not the partners' retained memories or signed receipts.

## Understand the result

- **Reply received:** the service reports a model reply. Review it.
- **Service reply:** text without a reported model invocation; it may be a refusal or a request to clarify.
- **Unavailable:** the local service is busy, warming or unavailable. No completion is established.
- **Unknown:** a response could not be confirmed. The request may have run. Do not automatically retry.

**Stop waiting** ends this page's wait. It does not cancel a request already received by the server. Closing a tab does not establish cancellation either. Refreshing a pending request changes its displayed state to unknown and does not resend it. The page keeps up to eight records per tab; export them before starting a separate tab if you need more.

## Continue a finite task on your computer

**Export a local workshop task** downloads a definition for the existing SPIRALMESH workshop: Claude drafts, Codex reviews, and Claude finishes. It includes your supplied material and, when available, the public reply labeled as unverified context. Downloading starts nothing. An unresolved request cannot be exported as a continuation task.

Use the [existing workshop guide](https://article11.ai/docs/spiralmesh-workshop.md) and [SPIRALMESH download](https://article11.ai/fork/#spiralmesh-plugin) for prerequisites and the `workshop create` / `workshop next` steps. The exported file is the task-definition input to that existing workflow. This is not a hosted queue or an automatic connection to Steve's household.

Running the local workshop is a separate decision: its configured Claude/Codex clients use your subscription allowance and permitted local libraries, and provider prompts may include stage context and permitted operation results. No API billing fallback is added by this page.

## For agent clients

Read the [machine-readable workbench contract](https://article11.ai/docs/public-workbench.json) and existing [API guide](https://article11.ai/api). The supported request remains `POST /api/v1/query` with `question` and optional `node`; this document does not introduce another endpoint or grant authority to call it.

Use only a task your operator has authorized, keep private material out, distinguish service notices from model replies, and preserve unknown outcomes. The existing [MCP rules reader](https://article11.ai/docs/mcp-quickstart.md) supplies documents and citations; reading a rule is not enforcement or permission to act.