Skip to content

Projects

Projects let you group knowledge around your active work. Instead of manually tagging every bookmark, Knowmarks uses semantic matching to automatically suggest relevant items.

Creating a Project

From the dashboard

Click Projects in the sidebar, then + New Project. Provide:

  • Name — A short identifier (e.g., "API Redesign")
  • Context document — Free-form markdown describing the project. The more detail you provide, the better the auto-association works.

From the CLI

km project add "API Redesign" -d "Redesigning our REST API to use GraphQL"

Via MCP

create_project("API Redesign", context_doc="Redesigning our REST API to use GraphQL...")

How Auto-Association Works

When you create or update a project, Knowmarks scans your collection for relevant items using multi-facet scoring:

  1. Context document is analyzed to extract search facets (key topics and concepts)
  2. Each facet generates a separate vector search against your collection
  3. Items are scored across all facets using a score-gap algorithm (mean + 2 sigma) that adapts to the distribution — specific projects get fewer, more precise matches; broad projects get more
  4. Items above the gap threshold are auto-associated

This means a project about "Kubernetes migration" will find your saved articles about container orchestration, Helm charts, and cloud infrastructure — even if they never mention "Kubernetes" directly.

Pin and Dismiss

  • Pin — Manually add an item to a project. Pinned items stay associated permanently.
  • Dismiss — Remove a suggested item. Dismissed items won't be re-suggested.

From the dashboard, use the approve/dismiss buttons on suggested items in the project view. From the CLI or MCP, use add_to_project and remove_from_project.

Project Metadata

Projects support structured metadata that improves matching:

  • Keywords — Comma-separated terms for keyword matching
  • Goals — What you're trying to achieve
  • Tech stack — Technologies involved

You can provide a context document (recommended) or structured metadata, or both.

Refinement

When an LLM provider is configured, you can refine a project's associations:

km project refine "API Redesign" --feedback "less generic tutorials, more implementation patterns"

This re-ranks borderline items using LLM judgment and records your feedback for future scans.

Prospective Notifications

When new items are saved that match an existing project's criteria, Knowmarks flags them in Pulse. Check the "Project activity" section to see which projects have new relevant content.