llms.txt
Compare with other sites' llms.txt files →# ScreenCI
> Tutorial and product videos made by your own coding agent: recorded on your machine, watched and edited in the browser, optionally exported as Playwright tests for CI
## Overview
ScreenCI is the tutorial and product video tool for SaaS teams: help-center walkthroughs, onboarding videos, docs screenshots, and launch clips. Anyone on the team, not just the person with the screen recorder, describes the video they want in one sentence, pastes one prompt into a coding agent (Claude Code, Cursor, Codex, a desktop agent, or any harness that can run a shell command), and gets back a narrated, auto-zoomed, subtitled walkthrough of the real app. Recording happens on the person's own machine (or, later, in the team's own CI), never in a vendor cloud, so no login to the product is handed over. The finished video opens in the browser, where marketing and support edit narration, voices, overlays, and languages. Optionally, engineers keep the videos in the repository, where each one is a Playwright E2E test that CI re-records on every push, so a video never goes stale and a broken flow fails the run before customers see it. One video becomes 79 localized versions, recorded on the localized page.
The workflow is preview-first: `screenci preview` records a free live preview and opens the video in the web editor, and `screenci export` (or the Export button) renders the finished videos and downloads them. The CLI runs everything that touches the app locally, so what leaves the machine is the raw recording plus timing data, never the app's source code. Exports are rendered in the cloud and delivered from a CDN. The CLI is open source. Trying it is free and needs no account.
## Key Concepts
### Everything by prompt
- Every button in the web app (Add project, Add video, Add screenshot, Edit, Add a language, Record all / Re-record, Add to CI) produces a one-line prompt with a one-time setup code
- The person types one sentence (what the video should show, or what should change); the prompt, the code, and the brief are generated
- The agent runs `npx screenci@latest setup <code>`, writes or edits the script, records the live preview, and the browser tab that made the prompt opens the video
- AI context (repository, site URL, sign-in, notes) and Branding (look, size, cursor, voice) are set once for the organisation and every prompt starts from them
### Edit in the browser
- Anyone in the organisation rewrites narration, swaps voices, moves overlays, cuts sections, adds languages, and changes render options in the web editor, with no prompt and no code
- Edits apply to the preview and every later export right away; edits that change what is captured are badged "applies after next recording"
- Every plan includes the editor for every member
### Videos as E2E tests
- Under the hood each video is a Playwright-based `.screenci.ts` script
- The scripts can live next to the product's code or only in ScreenCI (every preview uploads its sources, and every version keeps the sources it was recorded from, so Edit and Re-record start from the version being viewed); Add to CI stores a project-scoped key in the CI provider and adds a pipeline that records on every push
- Teammates with and without the repository share one video: outside the repository the agent points that video at the live site (`video.use({ baseURL })`) when the scripts name a dev server it cannot start, each preview lands as its own version, and the person picks the version that serves
- A flow that broke fails the run like any E2E test; a UI change re-records the video; the public URL keeps serving the selected version until a new one is selected
### Local record, cloud render
- The open-source CLI records locally; only the raw recording and timing data leave the machine
- `preview` is free and needs no account; `export` requires a paid plan, renders in the service, downloads, and delivers from a CDN
- Public URLs and embeds for published videos; `export --select` serves the finished render at the video's public URL
## Documentation
### Start here
- Overview (how ScreenCI works for the whole team): /docs
- Make Videos by Prompt (every button, no repository needed): /docs/make-videos
- Prompt Cookbook (example descriptions per button): /docs/prompt-cookbook
### For the team
- Narration and Voices: /docs/guides/narration
- Languages: /docs/guides/languages
- Overlays: /docs/guides/overlays
- Branding: /docs/guides/branding
- AI Context: /docs/guides/ai-context
- Screenshots: /docs/guides/screenshots
- Public URLs and Embeds: /docs/guides/public-urls-and-embeds
- Version History: /docs/guides/version-history
- Organisation and SSO: /docs/guides/organisation
- Signing In to Your App: /docs/guides/signing-in
- Free Trial Without an Account: /docs/guides/anonymous-trial
### Code and CI
- Repository and CI (export prompted videos as E2E tests, Add to CI, failed runs): /docs/repository-and-ci
- Start in a Repository (the integration prompt): /docs/agent-integration
- Manual Setup: /docs/manual-setup
- Video Script Basics: /docs/video-script-basics
- CI Setup: /docs/ci-setup
- Animated Interactions: /docs/guides/animated-interactions
- Camera and Zooming: /docs/guides/camera-and-zooming
- Mid-Video Overlay Updates: /docs/guides/overlay-updates
- Redact Sensitive Content: /docs/guides/redact
- Screen Audio: /docs/guides/screen-audio
- Update ScreenCI: /docs/guides/update-screenci
### Reference
- CLI: /docs/reference/cli
- Configuration: /docs/reference/configuration
- Public Delivery API: /docs/reference/public-delivery-api
- Full API Reference: /docs/reference/api
## AI Integration
### Setup codes from the web app
A signed-in member clicks a button in the ScreenCI app and receives a prompt that carries a one-time setup code and points at the brief for that button, for example:
Create a product video with ScreenCI by fetching https://screenci.com/add-project.md and following its steps with setup code SC-XXXX-XXXX. The video should show: '<description>'
The seven briefs, one per button, are agent-readable pages:
- /add-project.md: a new project and its first video
- /add-video.md: another video in an existing project
- /add-screenshot.md: a branded still from the same scripts
- /edit-video.md: change an existing video or screenshot
- /add-language.md: translate the narration and record one more language
- /record-video.md: re-record one video or every video with the scripts as they are (through the CI pipeline when the project records there)
- /add-to-ci.md: store a CI key in the provider, add a pipeline that records on every push, and trigger the first run (Add to CI)
`npx screenci@latest setup <code>` exchanges the code for a project-scoped `SCREENCI_SECRET` (written to `screenci/.env`, the only credential), uses the `./screenci` workspace when one exists, else pulls the project's current scripts from ScreenCI or scaffolds a new project (cloning the repository from AI context as read-only context when the command runs outside it), installs the agent skill, and prints a brief with the task, the site, how to sign in, the branding, the team's notes, and the commands to run: `npx screenci test`, then `npx screenci preview "<title>"`. Previews and exports upload a snapshot of the folder's scripts, and the browser tab that created the prompt opens the resulting video. Details: /docs/make-videos
### Working with the person
The person who sent the prompt is often a teammate who does not code (a marketer, a support lead, a docs writer) and may not use a terminal. Every brief and the installed skill tell the agent to:
- Never ask them to run commands, open files, or read the script
- Report in plain language: what the video shows, what changed, what needs their attention; no selectors, file paths, or command output unless asked
- Say exactly what to click when they are needed (the sign-in card in the browser the agent opened) and wait
- Never ask for a password, a one-time code, or an API key; `screenci login` is the only sign-in path, and the person signs in themselves in a real browser
- Finish the final message with the video link that `preview` printed (or the pipeline run link) on its own last line
- When a video records against the live production site, stop and ask the person before any step that acts on the real world (placing an order, paying, sending an email or invite, deleting or publishing something, changing account or billing settings) and wait for the answer; a dev, staging, or test deployment is safe to act on freely
- Never add `[pronounce: ...]` tags on its own initiative; only when the person reports a word is said wrong
- Deliver the result the way the brief says (a live preview, a pipeline run, or a pull request) and never switch paths because the repository has CI: only the codes that ask for a pipeline run complete on one
### Integrating ScreenCI from inside a repository
The canonical, agent-readable integration brief is at /integrate.md (npm), with /integrate-pnpm.md and /integrate-yarn.md for other package managers. Point a coding agent at the one matching the project's package manager:
Integrate ScreenCI by fetching https://screenci.com/integrate.md and following its steps, then add a screenci video about '<feature or files>'
The brief tells the agent to scaffold non-interactively (`npm init screenci@latest -- --yes`, which also installs the `screenci` agent skill at the repository root), author a `.screenci.ts` for the requested flow, then run `npx screenci preview "<title>"` (it runs the script itself) to record the free live preview and print the video link. `preview` needs no account (it runs under a free anonymous trial session, multi-language previews included); `npx screenci export` is for finished videos and requires a paid plan. Details: /docs/agent-integration
For a live site without a repository, /integrate-live.md is a compact brief: `npm init screenci@latest -- --yes --brief` in any folder, follow what it prints, explore with `npx screenci explore <url>`, then preview.
### Markdown Access
Every documentation page is also available as raw markdown by appending `.md` to its URL, for example /docs.md (the overview), /docs/make-videos.md, /docs/repository-and-ci.md, /docs/video-script-basics.md, or /docs/guides/camera-and-zooming.md.
## Common Use Cases
- **Launch videos that ship the same day as the feature**: product marketing types one sentence, no recording session.
- **Onboarding and help-center videos that never go stale**: committed videos are E2E tests; a UI change re-records them or fails the run.
- **Every market, every language**: one video becomes 79 localized versions with AI narration and subtitles, recorded on the localized page.
- **Docs screenshots that update themselves**: branded stills from the same scripts, embedded via public URLs that always resolve to the selected version.
- **No tickets in either direction**: marketing never files a ticket to engineering for a video; engineering never records one.
## Technical Details
- Browser automation and stable selectors via Playwright.
- AI narration synced automatically to visible UI actions, with subtitles.
- Local recording, cloud rendering, CDN delivery, and public URLs/embeds.
- CI/CD integration through a generated GitHub Actions workflow (templates for GitLab CI, CircleCI, Buildkite, and shell), versioned in Git; one-click Record all through the ScreenCI GitHub App.
## Comparisons
- https://screenci.com/compare/loom-and-screen-studio: ScreenCI against manual screen recorders. They are better for one-off, personal, spoken-in-your-own-voice videos; ScreenCI is for published walkthroughs that someone would otherwise have to record again after every UI change.
- https://screenci.com/compare/arcade-and-storylane: ScreenCI against interactive demo tours. Tours replay captured screens and are the right answer for a self-serve, clickable sales demo; ScreenCI records the real product running and returns an embeddable narrated video.
## Getting Help
- Documentation: https://screenci.com/docs/
- GitHub: https://github.com/screenci/screenci
- Support: support@screenci.com
---
This file follows the llms.txt specification for AI-readable documentation.
Learn more at: https://llmstxt.org/
Last updated: 2026-09-16