Skip to content
Community open-source Web · 0.2.7 release docs · 0.2.7Content reviewed: 2026-10-22View Markdown

3.14 Organize work experience automatically ​

Auto Dream turns an employee's work records into daily summaries and automatically saves reusable experience. You can inspect the evidence and withdraw memories that no longer apply. You do not need to approve each memory before it takes effect.

When it runs ​

Auto Dream is on by default and has no on/off switch. In Auto Dream in Settings you can change the daily generation time and time zone (daily at 23:30 in Asia/Singapore by default; existing schedules use their saved values) and see the last 7 days.

When an employee becomes active, a summary schedule is created for it. If employees that were active before an upgrade still have no Auto Dream plan, the page says so; open that employee's profile, pause them, then resume them to create the missing plan.

The local service must be running. Summaries cannot be generated while the computer is asleep or shut down, or after the service exits.

Auto Dream scheduling and digest directory. Check history against actual source records.

0.2.4 actual interface with example data. Auto Dream scheduling and digest directory. Check history against actual source records.

View summaries and memories ​

  1. Open Auto Dream settings to review recent records and generation status.
  2. Open the relevant employee to view daily summaries and the linked work evidence.
  3. In the employee profile, open Edit > What it remembers to review automatically saved experience and its sources. Withdraw anything that does not apply.

Without enough work evidence, do not expect Auto Dream to generate experience from nothing.

How saved experience is used ​

In later tasks, an employee can refer to active employee memories whose sources remain valid and whose business scope matches the task. Saving a memory does not mean every employee or conversation will use it.

The current implementation selects the most recent eligible employee memories, up to 30 per run. They provide context to the model; they do not guarantee the same judgment every time.

Withdraw a memory ​

Open the employee's memories, check the source, and withdraw the memory. A withdrawn memory should not become active again when the same source is retried.

Withdrawal affects future use. It does not rewrite completed task results. Content whose source is no longer valid should also stop being used as valid evidence.

Employee memories and your digital twin ​

ContentPurposeCurrent boundary
Employee memoriesSave reusable experience for that employeeAvailable to later tasks according to source, status, and business scope
Digital twin habit observationsCapture personal judgments and preferences from your actionsAutomatically saved as private samples; not yet automatically added to later ordinary conversations

The twin’s habit-notes empty state. No digest exists here; an empty record is not evidence of successful execution.

0.2.4 actual interface with example data. The twin’s habit-notes empty state. No digest exists here; an empty record is not evidence of successful execution.

Here, “distilling experience” means organizing it into configuration, skills, or memories. It does not involve training model weights.

Troubleshoot missing summaries ​

SymptomWhat to check
The page says an employee has no Auto Dream planOpen that employee's profile, pause them, then resume them to create it
Nothing was generated at the scheduled timeWhether the service was running, the schedule was enabled, and the time and time zone were correct
No evidence is availableWhether there were work records to summarize that day
A saved memory was not usedWhether it was withdrawn, its source is valid, and the task's business scope matches
The digital twin did not affect the next chatPrivate samples are not yet automatically added to ordinary conversations; this is a known boundary

Before triggering generation again, review existing summaries and run results. Distinguish missing evidence from generation failures and page loading failures.