Security
Your AI, your data, your call.
Launchcrier reads what it needs to write accurate posts, runs AI on your own account, and never shares your content with anyone.
Your AI keys
- You connect your own AI provider: Claude, ChatGPT (OpenAI), Gemini, a Google Cloud project, or any OpenAI-compatible model.
- Keys are encrypted (AES-256-GCM) before they are stored, with the encryption key held in a managed secret store. After you save a key, the app only ever shows its last four characters.
- Your key is used only for your workspace’s work. We never use it for our own purposes, and we never fall back to a key of our own: if your provider is unavailable, jobs pause and we tell you.
- We require paid-tier Gemini keys, because free-tier data may be used by the provider to improve its products.
What runs where
| Work | Runs on |
|---|---|
| Writing, planning, chat and reading images | Your AI provider, with your key |
| Search over large catalogs and document sets | Your AI provider, when it offers it; otherwise keyword search on our side |
| Checking each claim against its source | Your AI provider — ideally a different model than the writer |
| Images and video | Your AI provider, when it offers them; otherwise templates and your uploads |
| Exact checks on numbers, prices, dates, internal names and secrets | Launchcrier, without AI |
| Extracting text from files, keyword search, connected-learning statistics | Launchcrier, without AI |
Your sources
- You choose what Launchcrier can read: files, text, images, your company profile, your website, your store catalog, and optionally code repositories and design files.
- Repository access is a GitHub App you install on the repositories you choose, with read-only permissions (Contents and Metadata). Launchcrier keeps plain-language facts with links to their source, not a copy of your codebase.
- Everything you connect is treated as data, never as instructions, so text like “ignore your rules” in a file or web page is not obeyed. Writing steps have no tools that act on the world; publishing happens only from the approval inbox.
Your workspace
- Every request is scoped to the signed-in member’s workspace; workspaces are isolated from each other.
- Within a workspace, each brand’s data stays with that brand. A prompt only ever contains one brand’s facts, brief, tuning and metrics.
- Browsers never read the database directly; all access goes through the server, which checks your session on every request.
- The app has no third-party analytics or advertising scripts.
- Data is stored in the United States. Launchcrier is available to US customers first; EU and UK signups will open later.
Connected learning
Connected learning turns anonymous statistics from many workspaces into playbooks that help every brand. It follows one rule: features in, rules out.
- What goes in: a fixed list of numbers and labels about each published post — platform, format, length range, posting hour and weekday, whether the draft was approved, edited or rejected, and engagement compared with the brand’s own usual — plus coarse labels such as company type, industry, audience size and region. Anything outside that list is rejected before it’s stored.
- What never goes in: post text, captions, files, images, code, catalog data, brand or product names, handles, links or audience data.
- How it’s computed: by plain code, not by calling your AI. It runs in a separate system with its own access, which can’t read workspaces, and workspace identities are replaced with a keyed code so nobody analyzing the data sees which workspace contributed.
- What comes out: rules that apply only when at least 10 companies are behind them, shown as typical ranges rather than any single company’s numbers, each with its evidence and a switch to turn it off for your brand. No other customer’s content is ever placed in your prompts.
- Your choice: it’s on by default for new workspaces and shown during signup. Owners can turn it off in one click for the whole workspace or exclude a single brand. New contributions stop immediately, pending records are deleted within 30 days, and you still receive every playbook.
Publishing
- Nothing is published without approval unless you turn on routine auto-posts for a platform.
- Product claims, pricing, anything controversial and replies to comments or DMs are never automatic.
- Every approval is recorded — who approved, when, and what was edited — so you have a record of human review.
- The checker blocks internal names, endpoints, file paths and secret-looking strings from public posts.
Subprocessors
| Service | Purpose |
|---|---|
| Google Cloud / Firebase (United States) | Hosting, authentication, database, file storage and background jobs |
| Your AI provider | AI work for your workspace, through the key you connect (Anthropic, OpenAI, Google or another provider you choose) |
| Your chosen publisher | Publishing to your social accounts, through the account you connect |
| Sources you connect | Reading what you choose, through the access you grant (for example GitHub, Figma or your store) |
Your AI provider, publisher and connected sources work under your own accounts with them. [Final sub-processor wording pending legal review.]
Report a vulnerability
Email [CONTACT EMAIL]. Techoship LLC reviews every report.
Launch it. We’ll announce it.
Connect your AI and one source — a file, your website or your store — and see your first drafts. 14-day trial, your own AI.