bolt.diy vs Bolt.new

Bolt.new gives beginners a structured on-ramp: a free tier capped at 300,000 tokens per day (1,000,000 tokens per month) to learn on, plus Figma and screenshot import so a wrong result can be fixed without touching code. bolt.diy matches on autonomous agent mode, but it has no vendor plan at all, no free tier to try before paying, and no screenshot import on record. It expects the builder to run the project locally and bring their own OpenAI API key from day one, a steep first step for someone who has never shipped an app.

For developers who want the AI to scaffold and then step aside, both tools deliver a full project code export, so that part is tied. Bolt.new adds bidirectional GitHub sync, keeping edits made in the editor and changes pushed from a local clone in step, the hard filter this audience needs and one bolt.diy does not publish. Both let you self-hostself-host - Run the software on your own servers instead of the vendor’s cloud. the finished output. On the backend, bolt.diy documents a bring-your-own Supabase integration while Bolt.new claims a native Supabase integration. With code export tied and only Bolt.new confirming real GitHub sync, it takes the win.

Internal tools teams need role-gated auth and a real path to reach company systems. Both tools rely on a bring-your-own Supabase backend for data, so that part is a wash. Beyond that, Bolt.new publishes team-relevant facts that bolt.diy does not: SSO on its Enterprise plan and built-in collaboration and multiplayer for teams working the same admin panel together, even though auth itself is not built in and instead depends on the Supabase integration. bolt.diy, run locally by one developer with no vendor plan at all, has no equivalent SSO or collaboration facts on record, making it a weaker fit for an ops or IT team building shared internal dashboards.

For builders targeting a phone rather than a browser tab, bolt.diy is the only one of the two with a documented mobile output path: a native Expo App Creation feature plus an Electron desktop app option. Bolt.new's stack is described as full-stack JS/TS web projects, with no native mobile output on record. Both let you self-hostself-host - Run the software on your own servers instead of the vendor’s cloud. the generated output and both run in an autonomous agent mode, and both document a bring-your-own Supabase backend, but those only matter once the core output format is right, and only bolt.diy actually produces native mobile apps rather than a web app.

Builders who want to run the app builder itself, not just its output, need bolt.diy: it runs offline as a local-first tool, ships under the MIT license, and bills through your own OpenAI API key rather than a vendor subscription. Bolt.new has no equivalent facts on record for running the builder itself locally or under an open license; it is a hosted product where self-hostingself-host - Run the software on your own servers instead of the vendor’s cloud. only applies to the generated output, which both tools support. For anyone who specifically wants the builder on their own machine or server with no vendor credit meter, bolt.diy is the only one of the two built for that.

Bolt.new documents Figma and screenshot import, letting a design-led team go from a file or screenshot straight to a working front end, a capability bolt.diy has no fact for. Bolt.new also backs that with versioning and rollback and multiplayer collaboration, so a team can iterate on the same import together and step back if a change misses the mark. bolt.diy's output stack adapts to whatever framework the prompt requests, which is flexible but does not address the design-import gap. Since the heaviest requirement for this use case, translating a design file into code, is only backed by facts on Bolt.new's side, it wins.