Skip to content

continuous-research

Run recurring research cycles against configurable domain configs. Each cycle discovers new entries, scores them through a weighted quality gate, compares against the previous knowledge base, and writes an updated KB with a change delta.

  • Area: Research
  • Category: research
  • Level: advanced
  • Input: conversation
  • Output: document
  • Setup: api-key
  • Roles: Head of Marketing, Product Manager

Runs recurring research cycles in OpenAI, DeepSeek, Perplexity, or Tavily and maintains a scored knowledge base.

How you ask

  • “Run a research cycle for the procurement-ai domain”
  • “Set up a new research domain to monitor competitor AI tools in logistics”
  • “What's new in the procurement-ai knowledge base since last cycle?”
  • “Run all overdue research domains”
  • “Generate a PDF report from the latest procurement-ai KB”

Inputs

  • domain-configs — Research domain configs — the cycles available to run.
  • domain_id — Research domain to run, matching a config in the research-domains folder (resource.research-domains)
  • provider — Research backend for this cycle (deepseek, openai, notebooklm, perplexity, tavily). Omit to use the provider recorded on the research-domain record. (deepseek | notebooklm | openai | perplexity | tavily)
  • domain-id — loaded from your vault context
  • domain-id-changelog — loaded from your vault context
  • domain-id-kb-latest — loaded from your vault context
  • schedule_filter — Restrict the sweep to domains matching this schedule (e.g. Monday, weekly)
  • topic — Ad-hoc topic or question to scan, without creating a persistent domain
  • domain_id — Domain whose existing KB entries are to be rescored
  • scoring_weights — Override weights per scoring dimension; defaults to the domain config
  • domain_id — Domain to report on
  • cycle_date — Cycle to report; defaults to the most recent
  • name — Human-readable domain name
  • scope — What the domain covers and what it deliberately excludes
  • provider — Provider used to run the research cycle for this domain (deepseek | notebooklm | openai | perplexity | tavily)
  • schedule — Cadence for the recurring cycle (e.g. Monday 08:00)

Outputs

  • Config changes found during the cycle (new search queries), written instead of editing the config; applied at Step 1 of the next cycle.
  • Dated knowledge-base snapshot for the cycle, kept beside the overwritten _kb_latest.md.
  • Per-domain changelog the cycle appends to. Read as context and written on every run.
  • Query set written before the provider call.
  • Raw findings; derive topic-slug by kebab-casing the topic input.
  • domain-config (markdown)
  • kb-latest (markdown)
  • personal-research (markdown)
  • prefer a permanent directory)
  • reranked-results (markdown)
  • research-all-report (markdown)
  • research-cycle-report (markdown)
  • research-report-pdf (pdf)
  • {input.root}/{domain-id}cycle.md
  • {me.inbox}/research_{domain-id}_{YYYY-MM-DD}.md
  • {resource.research-domains}/{domain-id}.md
  • {resource.research-kb}/all_domains_report_{YYYY-MM-DD}.md
  • {resource.research-kb}/{domain-id}/cycle_{YYYY-MM-DD}.md
  • {resource.research-kb}/{domain-id}/reranked_{YYYY-MM-DD}.md
  • {resource.research-kb}/{domain-id}/{domain-id}_changelog.md
  • {resource.research-kb}/{domain-id}/{domain-id}config_proposal.md
  • {resource.research-kb}/{domain-id}/{domain-id}_kb_latest.md
  • {resource.research-kb}/{domain-id}/{domain-id}kb.md
  • {resource.research-kb}/{domain-id}/{domain-id}report.pdf

Seen in these workflows