FluidDocs CLI · Publish live documents that answer and report back

What you get

What the document does
on its own.

Canva and Gamma made documents prettier. Nobody made them work after you hit send. Any document type you publish through FluidDocs keeps doing its job once the link is out of your hands.

Ask it anythingReaders question the doc. Answers grounded in its contents, in your voice.

Summaries on demandA reader pulls the short version without waiting on you.

Book the call in the docA reader can act on it without opening a new tab or starting a thread.

Always the latestEdit by prompt. The same link updates. Nothing to resend.

Reader analyticsWho opened it, how far they got, and what they asked.

Standard HTMLYours to export, extend, or take anywhere.

Common workflows

From a folder to a document
that answers questions.

Each one is a single command. Default output is a short human-readable summary; add --json the moment a script or an agent is doing the reading.

deploy.sh

Agent-native

Your coding agent
can run every command.

Flat verbs, predictable flags, and --json on every path, errors included. That is all a coding agent needs to deploy, edit and publish your docs without you in the loop.

A real session, shown in Claude Code

claude — ~/AI/upwork-examples/bantry

▐▛███▜▌ Claude Code ▝▜█████▛▘ ~/AI/upwork-examples/bantry ▘▘ ▝▝

>

? for shortcuts

Works with your other coding agents

fld is a shell command, so the same calls work anywhere your agent can run one.

Codex Cursor Windsurf MCP server

CLAUDE.md · AGENTS.md

## Publishing docs with FluidDocs Use the `fld` CLI. Always pass `--json`. Stdout and errors are both machine-readable. # find what already exists (add -R for nested folders) fld --json ls -R # create, update, publish fld --json create --name NAME --dir ./out fld --json write ID --file ./index.html --autocommit fld --json publish ID --slug SLUG # turn on Ask AI, grounded in the doc itself fld --json enable-chat ID --grounded # long jobs: add --wait to stream, or poll later fld --json chat --message "…" --project-id ID --wait fld --json poll JOB_ID # set this once and stop repeating --project-id export FLUIDDOCS_PROJECT_ID=…

Why it is safe to run unattended

--json everywhere, so agents never parse prose. Failures come back structured too

Flat namespace, no nested groups. fld --help lists every command in one place

Old command names stay as aliases, so a stale prompt doesn't break the run

Background jobs return an id. Stream with --wait, or reconnect with fld poll JOB_ID

fld ls is non-recursive by default, so an agent can't walk your whole org

fld track needs no login, so generated docs can report their own analytics

Full reference

Every command, in the docs.

Commands are flat, with no nested groups, and older names stay as aliases. Run fld --help for workflows, flags and examples in one place, or read the complete reference online.

CLI reference →

Docswhoami · ls · stat · cat · cp · templates

Deploy & draftscreate · redeploy · upload · write · patch · commit

Publish & AIpublish · vis · enable-chat · chat · deck-build · agent-task

Jobs & analyticspoll · result · track · analytics

Real documents