Skip to main content

​
About apps

Every app has three parts: views (the screens, built as React components), workflows (the logic, written as durable TypeScript functions that run reliably to completion) and models (where the data lives). Biscuit’s AI writes the code on Biscuit’s own framework, which is tied closely to the infrastructure. That lets the platform control what code can run and what it can reach.
Yes. Your company gets one workspace, each team gets its own space, and you decide who can use or edit every app. See Access control for how it fits together.
No. In Biscuit, you cannot see or edit the code of your app. However, Biscuit generates readable views from the code itself. Data models show as tables with their fields and types, and workflows show as diagrams of each step in the order it runs. See Workflows.
Apps run on managed cloud infrastructure operated by Biscuit.
No. The AI writes real TypeScript. It can use npm packages and call AI models. Biscuit shows workflows as step-by-step diagrams so non-developers can follow them, but what runs underneath is code.
Vibe-coding tools give you code and leave security, permissions and hosting to you. Biscuit’s AI builds on a framework and infrastructure designed together, so permissions, data protection and access control are built into every app rather than added by hand.

​
App data and databases

A database built for Biscuit, with SQLite underneath. Every workspace gets its own dedicated database. It works as both a regular database and a real-time sync engine, so when data changes, everyone using the app sees it live.
Apps never talk to the database directly. Each model has rules for who can read, create, update and delete records, and those rules can follow your team structure. If no rule is set, nobody can reach the data. See more in Access control.
No. Biscuit’s AI designs the data models as it builds, as small linked tables. You can guide it if you have a structure in mind, and change models any time in chat. You can see the database in Library → App → Tables.
Schema changes are always backwards compatible. When a field is no longer needed, it’s deprecated rather than removed, so there’s no migration that takes your database offline. If you’ve used Firebase, it works in a similar way.
Yes. Biscuit’s compiler analyzes every query in your app and creates the indexes it needs automatically. When a page loads, the data it needs is already included, so there’s no wait for a second round of queries.

​
Isolation between customers

Yes. Each workspace’s app data lives in its own single-tenant database. No two companies share a database. Account details such as users, teams and permissions are managed by a separate central service.

​
Integrations and credentials

API keys you connect (Stripe, email, third-party integrations) are stored encrypted by Biscuit. You enter these in a secure field, never in the chat. They’re encrypted per workspace, stored separately from your app, and can’t be viewed again once saved. The AI-written code never handles them; AI can never read the API keys as it is handled by Biscuit’s proxy layer. A key is only decrypted at the moment your app calls the outside service, through a proxy layer. You can rotate or remove keys in Settings.
No. The AI-written code never handles keys. When your app calls an outside service, the request goes through Biscuit’s proxy, which adds the key at that moment and only allows calls to approved domains. A key can’t be sent to an unknown server.
Yes. An integration is built like an app, so whoever installed it can ask Biscuit to extend it. For example, add a HubSpot API call the default integration doesn’t include. The credentials stay locked; only the integration’s logic changes.
You choose. Share the connection itself with another team so they can build with it. Or share only an app that uses it: people can use what the app does, but never get access to the credentials.

​
Access and governance

Yes, through edit access. Only people with Can edit access to an app can publish it, including to the web. Give Can edit only to the people who maintain each app, and keep sensitive tools in Private spaces or restricted access. Apps stay inside your workspace until someone with edit access makes them public. See Publishing your app.
Yes. Every workflow run is recorded, including who ran it and when. We also record what was published and who published it.
Not today. If this is a requirement for your company, email support@biscuit.so.

​
Biscuit AI security

Biscuit uses Large language models (LLMs) from several providers, including Anthropic, OpenAI and Google, plus open-weight models. Biscuit splits building into tasks, benchmarks models for each one and uses the best fit, so it isn’t locked into any single provider.
Apps can use AI for text, images and voice, for example a workflow step that writes a summary, generates an image or creates a voiceover. Biscuit benchmarks models for each kind of task and picks the best one for every step. You can switch models the workflow settings.Biscuit uses top models from frontier labs including OpenAI, Anthropic and Google, such as GPT-6, Claude Opus 5.5 and Gemini 4. New models are added as soon as they’re released. Any model that is on the frontier (price, quality or speed), and potentially useful for users of Biscuit is evaluated and added with the right description of where it can be used best.
No. Biscuit does not use your data to train AI models.
Today, Biscuit runs on infrastructure that includes the US. EU-only hosting is on the roadmap. If data residency matters for your company, email support@biscuit.so and we’ll talk through options.
Not yet, but both are in progress. Our SOC 2 Type 2 audit is underway, with ISO 27001 running alongside it. We partnered with Vanta and you can follow the audit status and see the controls already in place in our Trust Center.
A penetration test is underway as part of ISO 27001.