Continue interrupted tasks deliberately

Key takeaways

• Read the message first: connection, load, credits or local model?
• After a busy message, waiting a few minutes usually helps.
• Text already written is kept when the connection drops.
• Request missing sections in the same conversation.
• Split large tasks by chapter or topic.

Distinguish cancellation, failure and missing setup

The app shows running steps and error notices. Read the actual notice before using “Try again”. A missing AI connection needs setup; exhausted credits need a status check. Retrying the same source does not automatically make it readable.

After deliberate cancellation, inspect the retained draft to decide whether to refine it or start a smaller task. Do not treat processing as complete while completion and export are absent.

Start a controlled retry

1. Record the step and error.
2. Check licence, credits and actual AI connection.
3. Inspect sources and limit the task to one clear section.
4. Review the new result before requesting more sections.

Save results before restarting

Save any useful existing export separately. A screenshot or partial text is not a substitute for the actual file. Avoid changing sources and settings during a running task if you want a traceable review afterwards.

Choose the next attempt from the actual failure

Test a source issue with a small import, a missing AI connection with its connection test, and an ambiguous request with a narrower goal. These comparisons answer different questions. If small tests succeed, expand scope gradually. If the same small source fails, note its type and exact notice instead of launching more large tasks. Avoid overlapping retries whose results become difficult to distinguish.

A useful partial draft can remain as a working state but needs that label. Request missing sections with their actual source scope. Then review the assembled content for repetition and conflicting statements. The last step is inspecting the saved export. Only then is the task complete from the recipient’s perspective, even if text appeared in the app earlier.

The most common causes

Most interruptions have one of five causes: a broken internet connection, an AI that is temporarily busy at the provider, used-up credits, a local model that is not running, or a rejected key for your own AI. Docsi shows a separate message for each cause in your language.

Long documents take time: for remote services Docsi waits up to four minutes for further data and for local models up to 15 minutes, without a fixed overall limit. A long run time is therefore not automatically an error.

What the messages mean

“Busy” or “too many requests” means the AI provider is not accepting another request right now; on the web this is status code 429. Wait a few minutes and try again. “Key rejected” only concerns your own AI providers: check the key and your balance there.

“Local model unreachable” means Ollama or LM Studio is not running or no model is loaded. Start the program and load the model. If your credits are used up, Docsi AI pauses until your next billing month; local tools keep working.

Assessing partial results correctly

If the connection breaks after Docsi has already written text, he keeps that text and continues the task. If only part still arrives, check whether all requested sections are present and the source list is complete before saving the result.

A usable partial draft can stay as a working state. Ask for the missing sections in the same conversation instead of starting everything again; Docsi knows the current state and does not need to read the sources fully again.

Splitting large tasks

A summary of 400 pages in one go is more fragile than ten summaries of 40 pages each. Split large requests by chapter or topic, have an outline created first and then work out the sections one by one.

This has three advantages: an interruption costs less work, you can check each part right away and credit usage stays predictable. At the end, combine the parts in one document or have Docsi create an overall document from them.

When contacting support is worthwhile

Write to support@impulse-apps.de if the same message keeps appearing even with a small task and a simple file, or if Docsi does not start. Include your operating system, Docsi version, the time of the error and the exact message.

Do not send private files or license keys unless we explicitly ask for them. A description of the steps and a screenshot of the message without private content are usually enough to narrow things down.

A worked scenario

An example: you have Docsi create a report from a 200-page manual. After a few minutes a message says the AI service is busy. The part already written, two of five sections, is visible. You save it as a Word file and wait ten minutes.

Then you ask in the same conversation: “Write sections 3 to 5 following the same outline.” Docsi continues without reading the whole manual again. For the next large document you split the task into chapters from the start, so an interruption costs less work.

Checklist before sharing

After an interruption, a quick look at these points is worthwhile before you use the result. Otherwise an incomplete report is often noticed only by the recipient.

1. All requested sections are present.
2. The source list is complete and matches the markers.
3. Partial drafts are saved as a file.
4. The cause of the interruption is known before you restart.

Key terms explained

Interruption

A task ends before the result is complete, for example due to a lost connection or cancelling.

Status code 429

The HTTP response “Too Many Requests”: a service is not accepting further requests right now.

Partial draft

An incomplete result that can be reused as a working state.

Timeout

A service does not respond within the expected time; the connection is closed.

Local model server

A program such as Ollama or LM Studio that serves a language model on your computer.