Glossary
The terms Bonsai's documentation leans on, defined once. Where a concept has its own page, the entry links to it.
Artifact
One cached item: a single Markdown file with a YAML frontmatter header. Every fetch or import produces exactly one artifact.
Artifact type
What an artifact represents. A source artifact is a single page (fetched or imported for one URL). A research_note is a synthesis keyed by a topic and a set of source URLs rather than a single page. An index is a navigation hub or llms.txt site index. A section is a heading-level child extracted from a page. See Import synthesized research.
Cache key
The SHA-256 hash of an artifact's normalized URL (for a synthesis, a hash of its topic and sorted sources). The key is the artifact's filename and the reason two slightly different URLs reuse the same entry.
Capture method
How the content was obtained: static_fetch (plain HTTP GET, the default), browser_fallback (rendered in a headless browser, used for --rendered and some site modules), agent_supplied (handed in via import), route_markdown (the page's machine-readable .md route, when one is published), or github_source (raw source fetched from the page's GitHub edit link).
Compressed vs. detailed
The two variants stored for every artifact. compressed is pruned to fit a tight context window; detailed keeps the full content. One fetch produces both. Pick one with --format. See Compression.
Conditional request
A revalidation request that sends the stored ETag or Last-Modified so the server can answer 304 Not Modified without resending the body. This is what makes refreshing a stale entry cheap. See Caching & Freshness.
Extraction confidence
How sure Bonsai is that it captured the real article: high, medium, or low. A low value usually means the page rendered client-side or lacked a clean article structure. See Troubleshooting.
Freshness
The state of an entry relative to its tier or TTL, computed at lookup time: fresh, stale_grace (past fresh but inside the grace window), or stale_expired (past the grace window). See Caching & Freshness.
Grace window
The period after an entry stops being fresh during which Bonsai tries a cheap conditional request before a full refetch. Each tier sets its own grace window.
Normalized URL
A canonical form of the requested URL — lowercased scheme and host, sorted query parameters, no fragment, default ports stripped — computed before hashing so equivalent URLs map to one cache key. See Caching & Freshness.
Revalidation
Checking whether a stale entry is still current using a conditional request, rather than downloading the page again.
Site module
A small per-domain plug-in that overrides parts of the fetch pipeline for a site that needs special handling (such as JavaScript-rendered pages). See Site Modules.
Storage mode
Where the cache lives: global (machine-wide, the default) or project (committed under .bonsai/research/ in a repo). See Storage Modes.
Summary aggressiveness
How hard prose is condensed when structural compression alone leaves compressed close to detailed: conservative (default), balanced, or aggressive. Headings, code, tables, and lists are never condensed. See Compression.
Tier
A freshness preset that sets the fresh and grace windows: volatile (short), standard (default), or stable (long). Override the exact window with a TTL. See Caching & Freshness.
Token estimate
An approximate token count (≈ chars / 4) stored per variant, so an agent can choose compressed or detailed against its budget.
TTL
A custom freshness window (--ttl 2h, 7d, 30d) that replaces the tier's fresh window. The grace window is then derived proportionally from the active tier. See Caching & Freshness.