Discovery and Search

People find published work four ways: browsing, semantic search, trending and ordinary web search. This page covers each, and what makes your own folders findable.

Four ways in

Published folders and profiles are also in the sitemap, so a web search can land someone on your /s/<token> page directly. Files behind a share link appear in none of these.

Semantic search

Semantic search matches meaning. Ask for "how we onboard new contractors" and it returns a file that only ever says "new starter checklist", because the two mean nearly the same thing. A keyword search finds that file only if you guess one of its words.

How it works, in order:

1. Chunking. Monofile splits each file's text into passages of roughly 1,200 characters with 150 characters of overlap, and labels each passage with the file's title and path so that context travels with it.

2. Embedding. Every passage becomes a vector, a long list of numbers that positions the passage by meaning.

3. Querying. Monofile embeds your query the same way and returns the closest passages by cosine similarity, each with its score.

4. Filtering. Monofile drops matches below the default floor of 0.3 similarity, and a search returns at most 20 results.

Because ranking is by similarity, a full sentence usually works better as a query than two or three bare keywords.

Namespaces

Search runs against one of two sets of content, or both:

Keyword and exact search

When you know the exact string, use one of three exact tools:

• The community browser's filter matches folder names only.

• public_bash runs read-only grep, find, ls, tree and cat inside any published folder, which is the fastest way to answer "does this project mention X".

• The command palette searches your own workspace. See Finding Things.

Indexing and the delay

Indexing is automatic and runs in the background just after a write. In practice:

• Publishing a folder indexes everything in it, file by file. A large folder takes a moment to finish, so a brand new share may be incompletely searchable for a short while.

• Saving a file inside a published folder reindexes it, in both the personal and the community index.

• Unpublishing removes the community entries for that folder, so an unpublished folder stops appearing in community search.

• Unchanged content is not reindexed. Saving a file whose text did not change skips the index.

• Only text is indexed. An empty file, or a file whose content is an image or video, contributes nothing to search.

Making your work discoverable

• Publish at the top level. You can publish root folders only, and the whole subtree travels with them. See Public Folders.

• Write a README.md. It becomes the description on community cards, and the index treats it like any other file.

• Name files and folders in words a stranger would use. Every passage carries its title and path into the index, so a good path improves retrieval as much as good prose does.

• Trending decays. Views, forks, comments and votes all feed the hot score, but age pushes everything down. A folder's best chance at the trending list is its first day or two. See Community.

• Publish a SKILL.md if the folder is a skill for agents. That flags it as a skill in the community browser.

From an agent or the CLI

The semantic_search tool takes a natural language query, a namespace and a limit, and needs the search:semantic scope.

{"query":"retry policy for failed uploads","namespace":"both","limit":5}

monofile search "retry policy for failed uploads" --namespace community --limit 5

curl -X POST https://www.monofile.ai/api/tools/semantic_search \
  -H "Authorization: Bearer $MONOFILE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query":"retry policy for failed uploads","namespace":"both","limit":5}'

Each result carries a similarity score and the file it came from. An agent can use the score to decide whether a hit is worth a read call.

Next steps