> ## Documentation Index
> Fetch the complete documentation index at: https://docs.useorgx.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Pick up where you left off

> Recover a decision, its evidence, and what would change it in a fresh conversation. Try it with example data and no local installation.

Finish a conversation without losing the reasoning your next session needs.
In this exercise, you will save one decision, open a fresh conversation, and
recover the choice, rejected option, and reconsideration condition from cited
OrgX records. The useful outcome is less reconstruction before the next action.

[Explore the interactive example](https://useorgx.com/try) first, without an
account. That walkthrough uses prepared example records; it does not run MCP
or demonstrate retrieval from your workspace. The steps below test actual MCP.

Use OrgX MCP from a browser with made-up work. This path does not run the
wizard, install plugins or hooks, or change terminal or IDE configuration.
It still sends data to OrgX and your AI provider. It is not an offline sandbox.

## Before connecting

The first readback tests one benefit. The same work record can connect a wider
workflow: client context, recorded work and receipts, dependencies, decisions,
artifact review, and standards adopted from feedback. Explore the layer that
solves your next problem; you do not need to configure the whole system first.

<Accordion title="Explore the wider workflow when you are ready">
  \| Next need | Explore | What to inspect | | --- | --- | --- | | Use OrgX in
  your developer tools | [Wizard](/docs/guides/wizard-cli-onboarding) and
  [plugins](/docs/guides/openclaw-plugin-setup) | Supported clients, skills/rules,
  capture choices, and configuration changes | | Know what happened | [Work
  receipts](https://useorgx.com/agent-work-receipts) | Recorded action, source,
  status, output, and available external result | | Coordinate an outcome |
  [Worked
  initiative](https://useorgx.com/live/demo?mode=northstar\&state=pending) |
  Dependencies, eligible next work, and decisions that need a person | | Decide
  whether work is ready | [Proof room](https://useorgx.com/proof/demo) |
  Artifact version, checks, evidence, and human review | | Improve the next
  attempt | [Review-to-standard
  example](https://useorgx.com/live/demo?mode=melissa\&step=standard-adopted) |
  An explicit adopted standard and evidence in the later revision |
</Accordion>

These examples help you choose a workflow to evaluate. They do not establish
measured savings or guarantee that every client exposes every capability.

Use a personal device and accounts you are allowed to use for testing. Start
with an empty OrgX test workspace with no employer data or connected services.
If you cannot isolate an empty workspace, stop and ask
[hope@useorgx.com](mailto:hope@useorgx.com) for help before connecting.
Do not connect GitHub, Drive, Slack, local files, or other data sources.
Do not upload source code, logs, secrets, or real customer information.

## Connect in the browser

In Claude on the web, open **Customize > Connectors**, add a custom connector,
and enter `https://mcp.useorgx.com/mcp`. Connect through OrgX OAuth and review
the requested permissions. Team accounts may require an owner to add it first.
If your account cannot add a custom connector, stop; do not install the wizard
or upgrade a paid plan just for this test.

See [Claude's current instructions](https://support.claude.com/en/articles/11175166-get-started-with-custom-connectors-using-remote-mcp).
Enable only OrgX for this conversation. Keep tool approvals on and inspect
each write. A prompt is a test instruction, not a replacement for access controls.

## First conversation: store one conditional decision

```text theme={"dark"}
Use only OrgX. First call orgx_bootstrap and show the current workspace.
Do not write until I confirm that this is my empty test workspace.
Do not connect other services, inspect files, or start background work.

After confirmation, create one initiative named Continuity example.
Store this fictional decision and evidence as records in that initiative:
Choose queue A for a toy notification service because provider B is unavailable
in this fictional scenario. Reject queue C because it lacks delayed delivery.
This decision expires when provider B becomes available. The evidence is a
synthetic note, not a real benchmark or outage report.

Show the stored record IDs and read the records back. If a write or read fails,
report the error; do not claim the record exists. Stop after the readback.
```

Copy only the initiative ID for the second conversation. Check that the
rejected option, reason, and condition survived storage before continuing.

## Second conversation: test continuity

Start a fresh conversation with only OrgX enabled. Do not paste the earlier
decision or explanation. Replace the placeholder with the stored ID:

```text theme={"dark"}
Read OrgX initiative [initiative ID]. What did we choose, which alternative
did we reject, and what would make us reconsider? Cite the stored records.
Separate evidence from assumptions. Do not invent missing reasons or change
any records. If the records are unavailable, say so.
```

Pass: actual OrgX readback recovers the choice, rejected alternative, and
reconsideration condition. Fail: it guesses, loses the condition, or cannot
retrieve the records. A fluent answer alone is not a pass. Disable cross-chat
memory where available; only cited OrgX readback counts as evidence.

## Finish and give feedback

Before trying a real project, compare this with how you resume work today.
Measure time to the first correct next action, missing or invented facts, and
the corrections you make. Set your own acceptable bar before the test; this
example is not a measured productivity claim.

For a CTO evaluation, require both a useful readback and an acceptable data
boundary. Ask which tool arguments leave the client, which credentials grant
access, how to revoke them, and how stored records and provider copies are
removed. Review [Security & Privacy](/docs/resources/security). Unanswered retention,
residency, subprocessors, or access-control questions remain open requirements,
not reasons to assume a guarantee. Use example data until they are resolved.

Reply to your invitation with where you got stuck, whether the second answer
was useful, and what it missed. Redact identifiers in screenshots.
Review and delete only the synthetic trial records using authorized OrgX
tools, then disconnect the connector and revoke its credentials. Do not delete
an existing workspace indiscriminately.

Disconnecting does not erase stored records or your AI provider's chat history.
For workspace deletion or remaining-copy questions, contact support.
Read [Security & Privacy](/docs/resources/security) before trying real work data.