Bolt.new trains open-weight models on user sessions, paid teams can opt out
testingcatalog · x · 2026-09-15
According to testingcatalog, StackBlitz's Bolt.new uses a recurring consent screen to let users opt in to training open-weight models (in partnership with Arcee AI) on their sessions, with secrets and personal data stripped first.
- Scope: Standard and Max plan data is not used for training; Teams and Enterprise workspaces are excluded.
- Product: a new Forge agent now sits alongside Standard and Max in the picker; Bolt includes Forge at no extra cost on every individual Pro plan through October 14.
Bolt's site claims its agents auto-route to the right model, cut errors by 98%, and handle projects 1,000x larger.
More from coding & agent
- Astra Agent Builds a NYC World Then Plays In It Itself, At $0.40/Minute — Lianhuiq · 2026-09-15
- Inspo: An MCP Server That Feeds 800+ Design References to Coding Agents — nutlope · 2026-09-15
- Sharing a Local Dev Runtime Between TUI and Coding Agents via MCP — vladimirlobas · 2026-09-15
- A short guide to building an AI agent harness from scratch in three modules — omarsar0 · 2026-09-15
- Orchestra conductor's hours-long score arrangement done by GPT-6 Astra agent — doodlestein · 2026-09-15
- Runway MCP adds music generation to your agent workflow — tlakomy · 2026-09-15