Examples

These examples are intentionally small and map directly to the worked/ material used in release validation, plus the installed-package source flows used in release smoke.

SwarmVault graph workspace
The local graph workspace after compile, benchmark, and review data are available.

Managed Source Example

Use this when you want the shortest path from a public repo URL to a synced vault plus a first-run brief.

swarmvault init --obsidian
swarmvault source add https://github.com/karpathy/micrograd
swarmvault source list
swarmvault source reload --all
swarmvault graph serve

Look for:

  • state/sources.json with a github_repo entry
  • state/sources/<id>/ working state for the managed source
  • wiki/outputs/source-briefs/<id>.md
  • the usual wiki/graph/report.md, state/graph.json, and state/search.sqlite artifacts after sync

Code Repo Example

Use the tiny repo-shaped example when you want to exercise repo ingest, module pages, graph report generation, and benchmark output together.

swarmvault init --obsidian
swarmvault ingest ./worked/code-repo
swarmvault compile
swarmvault benchmark
swarmvault graph serve

Look for:

  • wiki/code/ module pages with local import links
  • state/code-index.json aliases and repo-relative paths
  • wiki/graph/report.md plus wiki/graph/report.json
  • state/benchmark.json

Research Capture Example

Use the capture workflow when you want normalized metadata and durable markdown before the source enters the standard compile path.

swarmvault init
swarmvault add https://arxiv.org/abs/2401.12345
swarmvault add 10.5555/example-doi
swarmvault add https://example.com/article
swarmvault compile
swarmvault query "What does the captured research say about trust signals?"

Look for:

  • source_type, canonical_url, authors, and publication metadata in the stored raw markdown
  • localized remote assets under raw/assets/ for article captures
  • compiled source pages in wiki/sources/
  • research-aware search and graph entries inside the local workspace UI

Review And Save-First Example

Use this loop when you want generated work to stay inspectable instead of mutating the vault silently.

swarmvault init
swarmvault ingest ./worked/mixed-corpus
swarmvault compile --approve
swarmvault review list
swarmvault review show <approval-id>
swarmvault query "What should I save as a report?" --format report
swarmvault candidate list

Look for:

  • staged compile changes in state/approvals/
  • candidate concept/entity pages in wiki/candidates/
  • saved report outputs under wiki/outputs/
  • refreshed graph/report artifacts after accepted changes

Book Reading Example

Use this when you want a chapter-by-chapter reading companion that compounds into a fan wiki instead of a pile of isolated notes.

swarmvault init --obsidian
cp ./worked/book-reading/raw/chapter-1.md raw/
swarmvault ingest raw/chapter-1.md
swarmvault compile
swarmvault query "What is the Thornback discrepancy and why does it matter?"

Look for:

  • wiki/entities/ pages for recurring characters like Maren Voss and Dr. Tomas Lune
  • wiki/concepts/ or related pages for themes, disputes, and settings that grow across chapters
  • denser backlinks and cross-references each time you ingest another chapter
  • query answers that synthesize plot threads across multiple chapters instead of one file at a time

Add chapter-2.md and chapter-3.md the same way to watch the wiki compound as the book progresses.

Research Deep-Dive Example

Use this when you want an evolving thesis wiki from overlapping papers and practitioner notes, with contradiction checks across claims.

swarmvault init --profile personal-research
swarmvault ingest ./worked/research-deep-dive/raw/rag-survey-2024.md --guide
swarmvault ingest ./worked/research-deep-dive/raw/compiled-knowledge-karpathy.md --guide
swarmvault ingest ./worked/research-deep-dive/raw/hybrid-approaches-2025.md --guide
swarmvault ingest ./worked/research-deep-dive/raw/production-rag-lessons.md --guide
swarmvault compile
swarmvault lint --deep

Look for:

  • thesis and concept pages that shift as new sources disagree or add nuance
  • source-linked claims under wiki/sources/ and concept/entity pages
  • contradiction findings from lint --deep when sources disagree on retrieval quality or hybrid approaches
  • guided-ingest artifacts that make each source addition inspectable before it reshapes the wiki

Personal Knowledge Base Example

Use this when you want a personal Memex that connects journals, health logs, podcast notes, and goals into one durable vault.

swarmvault init --obsidian --profile personal-research
swarmvault ingest ./worked/personal-knowledge-base/raw
swarmvault compile
swarmvault query "How does my sleep relate to my productivity?"
swarmvault graph serve

Look for:

  • source pages in wiki/sources/ for each journal, podcast, health, and goals entry
  • entity and concept pages that connect the same people, projects, and habits across source types
  • dashboards and graph links that make recurring patterns visible across otherwise unrelated notes
  • saved query outputs in wiki/outputs/ that cite across multiple personal sources

Source Material

The example inputs live in the repo so the docs and smoke tests stay aligned:

  • worked/code-repo/
  • worked/mixed-corpus/
  • worked/capture/
  • worked/book-reading/
  • worked/research-deep-dive/
  • worked/personal-knowledge-base/

If you want to start with the pattern before installing the CLI, use the standalone LLM Wiki schema template.

For real public-repo validation against the installed npm package, see OSS Validation.

When you refresh screenshots or docs examples, prefer running the installed CLI against those inputs rather than the workspace source tree directly.