Configuration Tips

Goal

This guide explains how to configure OctoReport in a way that is stable, cost-aware, and aligned with the current product spine:

Source → Library → Report → Delivery / Ask / Trigger Inbox

If you are just getting started, read these first:

• Product Overview
• Quick Start

1. Configure the main flow first

Recommended order:

1. Set up 1-2 reliable sources
2. Create 1 clearly scoped library
3. Run 1 report template with no more than 3 steps
4. Add Delivery or Trigger Inbox afterward

Do not start by configuring many sources, many libraries, and many templates at once.
That usually makes troubleshooting harder, not faster.

2. Source configuration tips

Search / RSS / Web scraping

• Start with stable sources and predictable structures
• Use narrower keywords for high-frequency sources to keep noise out of libraries
• Only enable heavier scraping and cleaning when it is actually needed

Deduplication strategy

• UPDATE

  : best for pages or feeds that keep changing

• KEEP_OLD

  : best for announcements, news, tenders, and event-like records

If you keep seeing repeated or near-duplicate content in a library, check deduplication before tuning prompts.

Schedule frequency

• Daily briefs: start with every 6-24 hours
• Monitoring workflows: start with every 1-4 hours
• Higher frequency only when there is a clear business reason

3. Library configuration tips

Libraries are valuable when their boundaries are clear.

Recommended:

• One library per clear theme
• Keep unrelated topics out of the same library
• Make source-to-library relationships easy to explain

Not recommended:

• A giant "everything" library
• Sending the same sources into many overlapping libraries

If you want long-term accumulation, first make the library a stable input layer for reports and Ask, then expand into long-term knowledge workflows.

4. Report template configuration tips

Start with a small template

Recommended structure:

1. Summarize
2. Analyze
3. Format output

Three steps or fewer are easier to validate and cheaper to run.

Model allocation

• Use smaller models for summarization
• Use more expensive models only for deep analysis
• Formatting can often go back to a smaller model

Prompt principles

• Be explicit about output format
• Be explicit about time range
• State that the answer should be based on provided content

If one step asks for summarization, evaluation, prediction, and table formatting all at once, it probably needs to be split.

5. Delivery configuration tips

Delivery should remain Email-first by default.

Recommended:

• Start with email delivery only
• Keep subject, recipients, and activation settings minimal
• Add Slack / Lark later only when needed

If your goal is simply to deliver reports reliably to yourself or a small team, you do not need multiple channels on day one.

6. Ask configuration tips

Ask works best for two things:

• Following up on content in a library
• Doing post-report clarification or expansion

To make Ask more reliable:

• Confirm the library actually contains relevant content
• Include time range and topic in the question
• Ask a specific question first, then expand with follow-ups

7. Trigger Inbox configuration tips

Trigger Inbox is an external trigger layer. It should not replace the main workspace setup.

Recommended:

• Configure sources, libraries, and report templates in the main workspace first
• Then expose Trigger Inbox as an external entry point
• Validate with email first, then decide whether Slack / Lark is worth adding

If Trigger Inbox has a low match rate, check:

• Whether template tags are clear
• Whether external triggering is enabled
• Whether requests include a clear topic and time range

8. A minimal stable setup

If you just want a practical baseline, start with:

• 2 sources: 1 RSS + 1 search source
• 1 library: single topic
• 1 report template: 3 steps
• 1 delivery: email
• 1 Ask workflow: follow-up on library content

This setup is the easiest to validate and extend.

Next

• Report Generation - Learn template design
• Library Management - Improve input-layer structure
• Optimization & Troubleshooting - Control cost and debug common issues