Agentic capture

Inkly turns a product session into a demo by capturing what a browser shows. The CLI opens Chrome, gives an agent a navigation harness, records each committed step, and writes a demo folder. There are two capture commands. They share the same start / stop / cancel surface and differ in what each step records.

Command	Records each step as	Best for
`inkly capture`	A screenshot, or a short video clip for steps with motion	Pixel-faithful walkthroughs, animations, and pages you only need to show
`inkly capture-html`	A self-contained, replayable HTML snapshot	Crisp text, resizable layouts, and snapshots that stay sharp at any zoom

Both commands print JSON for every subcommand. Neither command plans the demo story for you. An external agent drives the page through inkly capture nav.

How a capture session works

Each command exposes the same session lifecycle:

start — launch Chrome, open --url, and arm recording. Returns immediately with a session id.
status — inspect a running session and see recorded steps.
nav <verb> — drive the captured browser from an agent.
undo — drop the last recorded step before you stop.
stop — finish the session, build the demo, and write it to --out.
cancel — abandon the session and clean up.
profiles — list the persistent Chrome profiles the CLI knows about.
ask-user-to-log-in — open a headed login flow for a persistent profile.

start returns right away while recording continues in the background until you stop. Agents drive the browser with inkly capture nav or inkly capture-html nav.

inkly capture nav snapshot --session <id>
inkly capture nav click @0-5 --session <id>
inkly capture nav fill @0-8 "Acme" --session <id>
inkly capture nav wait load --session <id>

Use nav snapshot first. It returns an accessibility tree with stable refs like @0-5, which agents can use for clicks, fills, selections, uploads, and assertions. Common verbs include snapshot, refs, click, fill, type, press, select, upload, get, is, wait, back, forward, reload, scroll, and mouse.

inkly capture (screenshots and video)

Use inkly capture when you want to show exactly what the page renders. Still steps become screenshots. Scrolls and other motion become short webm clips with poster frames. Click zoom and pan are applied to screenshot steps by default.

# Start a session against a URL.
inkly capture start --url https://app.example.com --name "Onboarding tour"

# ... drive the page (or let an agent drive it) ...

# Finish and write the demo folder to ./out.
inkly capture stop --session <id> --out ./out

# Abandon a session instead of finishing it.
inkly capture cancel --session <id>

Key flags

Flag	Default	Purpose
`--url <url>`	—	URL to open and arm for recording (`start`).
`--name <name>`	—	Demo name written into the exported demo (`start`).
`--session <id>`	—	Session id from `start` (`stop` / `cancel`).
`--out <dir>`	—	Output directory for `stop`. Writes the ZIP and unpacks the demo.
`--video`, `--record-video`	on	Record motion segments. Use `--no-video` to disable.
`--zoom`	on	Apply click zoom/pan to screenshot steps. Use `--no-zoom` to disable.
`--width <n>`	`1440`	Browser viewport width.
`--height <n>`	`900`	Browser viewport height.
`--window-size <w>x<h>`	—	Shortcut for `--width <w> --height <h>`.
`--headed`	off (headless)	Use a visible Chrome window. Headless is the default.
`--browser <path>`	local Chrome	Chrome/Chromium binary to drive.
`--connect-to-browser <url>`	—	Attach to an existing Chrome DevTools endpoint.
`--profile <name>`	—	Reuse a persistent Chrome profile so a login survives across captures.
`--keep-profile`	off	Leave the temporary Chrome profile in place after `stop`/`cancel`.
`--timeout <ms>`	`120000`	Page wait timeout.
`--no-unpack`	off	On `stop`, write only the ZIP without unpacking the demo.

Inkly prefers real installed Google Chrome over Chrome for Testing, because bot-gated sites (Cloudflare, SSO providers) pass headed real Chrome but block the automation build. For an interactive login you need to watch, run the first capture with --headed.

inkly capture-html (self-contained HTML)

Use inkly capture-html when text crispness and layout matter. Each step is a self-contained HTML snapshot with the markup, styles, inlined assets, and scroll position needed to replay that screen. Because it is real HTML, it stays sharp at any zoom.

# Start an HTML capture session.
inkly capture-html start --url https://app.example.com --name "Dashboard walkthrough"

# Finish and write the demo to ./out, compacting media.
inkly capture-html stop --session <id> --out ./out

# Cancel a session.
inkly capture-html cancel --session <id>

inkly capture-html accepts every common capture flag, plus one HTML-specific flag:

Flag	Default	Purpose
`--block-videos`	off	Replace each `<video>` with its poster frame instead of inlining the video.

inkly capture is for screenshot and video steps. inkly capture-html is for HTML snapshots. Keep the two capture formats separate and choose the one you want before starting the session.

What a capture produces

stop --out <dir> writes a .zip and unpacks a demo folder beside it unless you pass --no-unpack. The folder is a standard Inkly demo:

<slug>/
  demo.config.json     the demo definition and its steps
  assets.json          maps each asset id to a file under public/
  public/<file>        captured image/video bytes, content-addressed by hash
  snapshots/snap-NNN/  captured HTML snapshots, when present

You can preview a bare exported demo folder directly:

inkly dev <slug>/

To add it to a hub, run inkly add --from from inside that hub:

inkly add <slug> --from ../captures/<slug>

How captures become assets

Captured media is stored as managed assets. Each binary lands in assets.json with a stable id, content type, and size. Inside demo.config.json, steps reference those entries with the asset:<id> URI scheme, such as asset:cap-001 for a screenshot. Video steps can also reference a poster asset. The runtime resolves each asset:<id> through assets.json. When you preview a hub locally, inkly dev serves the captured files from your machine. inkly sync is optional and helps keep large captured assets out of GitHub. inkly snapshot can publish one demo as a standalone link. See Publish a demo. For the manifest shape and the resolution rules, see Assets and Demo config.

Get started

Concepts

CLI

Chrome extension

Platform

Agentic capture

How a capture session works

Agent navigation

inkly capture (screenshots and video)

Key flags

inkly capture-html (self-contained HTML)

What a capture produces

How captures become assets

​How a capture session works

​Agent navigation

​inkly capture (screenshots and video)

​Key flags

​inkly capture-html (self-contained HTML)

​What a capture produces

​How captures become assets

How a capture session works

Agent navigation

inkly capture (screenshots and video)

Key flags

inkly capture-html (self-contained HTML)

What a capture produces

How captures become assets