Skip to content

customization.md

Latest commit

 

History

History
129 lines (99 loc) · 4.82 KB

customization.md

File metadata and controls

129 lines (99 loc) · 4.82 KB
Raw

Custom iRead subscriptions

The maintainer's example subscription tracks AI applications, embodied intelligence, and AI hardware together. The recommended path for another set of fields is to create a multi-domain manifest, run batch-propose, review the results, and run apply-subscription. The generated directory contains one database and one report stream for every approved domain.

Use separate configuration directories only when the user wants completely independent subscriptions, schedules, and databases:

mkdir -p profiles/my-field
cp config/accounts.json profiles/my-field/accounts.json
cp config/external_sources.json profiles/my-field/external_sources.json
cp config/profile.json profiles/my-field/profile.json
cp config/source_policy.json profiles/my-field/source_policy.json
cp config/topics.json profiles/my-field/topics.json
cp config/reporting.json profiles/my-field/reporting.json
cp config/entities.bootstrap.json profiles/my-field/entities.bootstrap.json

bin/iread --config-dir profiles/my-field init

Missing files fall back to the repository config/ directory, so a small profile can override only the files it needs.

Subscriptions created by apply-subscription contain subscription.json and runtime.json. The latter assigns separate data/profiles/<id> and logs/profiles/<id> directories so different subscriptions do not mix. Explicit IREAD_DATA_DIR and IREAD_LOGS_DIR environment variables override these values; the old REPORTER_* names remain compatibility aliases.

Target research fields

Edit profile.json first. This file describes the user intent that should remain stable even when the topic taxonomy changes:

  • name and description: the profile shown by the CLI, reader, and reports.
  • seed_keywords: the short industry or field description supplied by the user.
  • audiences: intended readers such as researchers, operators, or investors.
  • goals: material changes, trend detection, verification, or other decisions.
  • languages, regions, and exclusions: collection and analysis boundaries.
  • domains: the approved top-level fields and their second-level topic ids.

Edit topics.json to define the taxonomy used for enrichment and reporting:

  • topics[].name: the primary buckets shown in reports.
  • secondaries: subtopics used for trend aggregation.
  • keywords and cross_cutting_keywords: routing hints for classification.
  • event_keywords: domain-specific events such as financing, product launches, papers, policy updates, or hiring changes.

After changing topics, run a small enrichment batch before trusting the reports:

bin/iread --config-dir profiles/my-field enrich --max-batches 1
bin/iread --config-dir profiles/my-field report daily --no-publish

Target authors and feeds

Use accounts.json for WeChat public accounts:

{
  "name": "Example Author",
  "wechat_id": "example_author_id",
  "priority": "required",
  "aliases": ["Example"]
}

Use external_sources.json for RSS, Atom, newsletters, blogs, podcasts, and first-party publication feeds:

{
  "id": "example-author",
  "name": "Example Author",
  "priority": "required",
  "source_type": "independent_practitioner",
  "homepage_url": "https://example.com/",
  "feed_url": "https://example.com/feed.xml",
  "capture_method": "rss",
  "content_mode": "summary_or_link"
}

content_mode should describe what the upstream feed allows or provides. Prefer summary_or_link for feeds that do not clearly allow republication of full text.

After collecting and analyzing a sample, generate a source review:

bin/iread --config-dir profiles/my-field sources-review \
  --output data/my-field/source-review.json

The review separates domain fit, content quality, collection quality, and rating confidence. It also provides representative works for user confirmation. Rating weights and confidence thresholds live in source_policy.json.

Report strategies

The daily, weekly, and monthly sections in reporting.json support:

  • enabled: whether the scheduled report runs.
  • cadence fields such as publish_hour, weekday, and rolling window size.
  • reading_minutes: the intended reading budget.
  • focus: report goals such as new evidence, trend change, or prediction review.
  • max_articles: the maximum candidate set before event-level ranking.

The report prompt creates topic sections from topics.json; it no longer assumes an AI-specific report outline.

Local data separation

Profiles can use separate data and log directories through .env:

IREAD_CONFIG_DIR=profiles/my-field
IREAD_DATA_DIR=data/my-field
IREAD_LOGS_DIR=logs/my-field

This override is useful for custom deployment layouts. Generated profiles are already isolated through runtime.json; hand-built profiles should add that file or set these variables to avoid mixing article databases and reports.