flexkit knowledge
Manages the project’s knowledge: documents (PDF, Word, PowerPoint, Markdown, text, HTML) and web pages that the agent searches and cites. Flexkit extracts each document into passages and indexes them for semantic and keyword search. You can attach a document to a record, or to every record of an entity type, so the agent prefers it when it works on those records.
Synopsis
flexkit knowledge <subcommand> [arguments] [options]Running flexkit knowledge without a subcommand runs flexkit knowledge ls.
knowledge ls
List the documents you can read, newest first.
flexkit knowledge ls [--search <text>] [--status <status>] [--limit <n>] [--offset <n>] [--json]Examples
# List documents
flexkit knowledge ls
# Documents whose title or content mentions returns
flexkit knowledge ls --search "return policy"
# Ids of the documents that failed to index
flexkit knowledge ls --status failed --json | jq -r '.sources[].id'knowledge inspect
Show a document: type, size, status, passage count, who can read it, whether the agent can search it, the latest indexing job, and the records and entity types it is attached to. Alias: get.
flexkit knowledge inspect <sourceId> [--json]Examples
# Show a document
flexkit knowledge inspect 6b1f3c2e-4d5a-4e7b-8c9d-0a1b2c3d4e5f
# Print the indexing error, if any
flexkit knowledge inspect 6b1f3c2e-4d5a-4e7b-8c9d-0a1b2c3d4e5f --json | jq -r '.source.currentVersion.error'knowledge add
Add files, folders and https:// web pages. Files are uploaded straight to the Flexkit file store, then indexed in the background. Folders are walked recursively; hidden files and node_modules are ignored, and files of other types (images, spreadsheets…), empty files and files over 20 MB are skipped with a warning. Adding a file that is already in the project returns the existing document (already added). Alias: upload.
With --wait, the command follows indexing until every document is ready or failed, and exits with code 1 if any failed. With --link, every added document is attached to the given record or entity type.
flexkit knowledge add <file|dir|url...> [--title <title>] [--space <code>...] [--hide-from-agent] [--link <entity[:recordId]>...] [--concurrency <n>] [--wait] [--timeout <seconds>] [--json]Examples
# Add a PDF and wait until the agent can search it
flexkit knowledge add ./handbook.pdf --wait
# Add a folder of documents, five uploads at a time
flexkit knowledge add ./docs/policies --concurrency 5
# Add a web page with your own title
flexkit knowledge add https://example.com/shipping --title "Shipping policy"
# Attach a spec sheet to one product and share it with the sales space only
flexkit knowledge add ./chair-spec.pdf --link product:product-chair --space salesknowledge update
Change a document’s title, the spaces that can read it, or whether the agent can search it.
flexkit knowledge update <sourceId> [--title <title>] [--space <code>... | --all-members] [--hide-from-agent | --show-to-agent] [--json]Examples
# Rename a document
flexkit knowledge update 6b1f3c2e-4d5a-4e7b-8c9d-0a1b2c3d4e5f --title "Returns policy 2026"
# Only members of two spaces can read it
flexkit knowledge update 6b1f3c2e-4d5a-4e7b-8c9d-0a1b2c3d4e5f --space sales --space support
# Keep it for people, but out of the agent's search
flexkit knowledge update 6b1f3c2e-4d5a-4e7b-8c9d-0a1b2c3d4e5f --hide-from-agentknowledge rm
Delete one or more documents with their passages and links. The command asks for confirmation; pass --yes in scripts. Alias: remove.
flexkit knowledge rm <sourceId...> [--yes] [--json]Examples
# Delete a document
flexkit knowledge rm 6b1f3c2e-4d5a-4e7b-8c9d-0a1b2c3d4e5f
# Delete every document that failed to index
flexkit knowledge ls --status failed --json | jq -r '.sources[].id' | xargs flexkit knowledge rm --yesknowledge reingest
Index a document again: an uploaded file from its stored copy, a web page from a fresh fetch. The current version stays searchable until the new one is ready.
flexkit knowledge reingest <sourceId> [--wait] [--timeout <seconds>] [--json]Examples
# Re-fetch a web page that changed
flexkit knowledge reingest 6b1f3c2e-4d5a-4e7b-8c9d-0a1b2c3d4e5f
# Re-index and wait for the result
flexkit knowledge reingest 6b1f3c2e-4d5a-4e7b-8c9d-0a1b2c3d4e5f --waitknowledge download
Download the original file of an uploaded document, under its own file name. dest can be a file, a folder (existing, or ending in /) or - for stdout. Web pages have no stored file; open their address instead.
flexkit knowledge download <sourceId> [dest] [--force]Examples
# Download into the current folder
flexkit knowledge download 6b1f3c2e-4d5a-4e7b-8c9d-0a1b2c3d4e5f
# Download into a backup folder
flexkit knowledge download 6b1f3c2e-4d5a-4e7b-8c9d-0a1b2c3d4e5f ./backup/
# Stream the file to another tool
flexkit knowledge download 6b1f3c2e-4d5a-4e7b-8c9d-0a1b2c3d4e5f - | pdftotext - -knowledge search
Search the documents you can read the way the agent does, and print the best passages with the document title and location (page, slide, heading). Results include documents hidden from the agent, since you can read them.
flexkit knowledge search <query> [--limit <n>] [--source <sourceId>...] [--about <entity[:recordId]>] [--json]Examples
# Ask a question
flexkit knowledge search "how long do customers have to return an item?"
# Prefer the documents attached to one product
flexkit knowledge search "assembly instructions" --about product:product-chair
# Search two documents only
flexkit knowledge search "late fees" --source 6b1f3c2e-4d5a-4e7b-8c9d-0a1b2c3d4e5f --source 0c9d8e7f-6a5b-4c3d-2e1f-0a9b8c7d6e5f
# The text of the best passage
flexkit knowledge search "warranty" --limit 1 --json | jq -r '.hits[0].text'knowledge link and knowledge unlink
Attach a document to a record, or to every record of an entity type when you leave out the record id; detach it the same way. entity is the entity name from your schema.
flexkit knowledge link <sourceId> <entity> [recordId] [--json]
flexkit knowledge unlink <sourceId> <entity> [recordId] [--json]Examples
# Attach a manual to one product
flexkit knowledge link 6b1f3c2e-4d5a-4e7b-8c9d-0a1b2c3d4e5f product product-chair
# Attach a policy to every order
flexkit knowledge link 6b1f3c2e-4d5a-4e7b-8c9d-0a1b2c3d4e5f order
# Detach it from orders again
flexkit knowledge unlink 6b1f3c2e-4d5a-4e7b-8c9d-0a1b2c3d4e5f orderOptions
See also the global options.
Permissions and limits
- Reading (
ls,inspect,search,download, and following indexing with--wait) is open to every project member and every project API token. - Writing (
add,update,rm,reingest,link,unlink) needs the Owner or Developer role, for a person and for a project API token alike. - You only see documents shared with every member or with one of your spaces, and you can only change documents whose spaces you all belong to. A project API token sees the spaces it was granted when it was created.
- With a project API token, uploads count against the daily upload limit of the member who created the token.
- Files are limited to 20 MB. Your plan sets the number of documents per project and the uploads per member per day;
addreports the limit when you reach it.
Common workflows
# Load a folder of policies, attach them to every order, and wait until searchable
flexkit knowledge add ./policies --link order --wait
# Check that the agent will find the answer
flexkit knowledge search "can a customer cancel after shipping?"
# Re-index every web page, for example from a nightly job
flexkit knowledge ls --limit 100 --json \
| jq -r '.sources[] | select(.kind == "url") | .id' \
| xargs -n1 flexkit knowledge reingestThe agent also reads knowledge through the MCP server tools search_knowledge and read_source.