#

Key takeaways

• Bolt Forge is a research-preview agent inside Bolt.new.

• It uses open-source AI models inside the app-building platform.

• Pro plans get up to 50 times more Forge usage at no extra cost through October 14.

• Users can opt in to share anonymized build sessions for model training.

• Standard and Max sessions remain private.

• The reported score is company-provided, not an independent benchmark.

What Bolt announced

Bolt.new announced Bolt Forge on September 14, 2026. The company describes it as a research-preview agent. It runs open-source AI models inside the app-building platform.

The announcement frames Forge as part of the product, not as a separate service. That matters because the feature is tied to how users build inside Bolt.new. The source does not add more technical detail about how the agent works.

Usage changes for Pro plans

Bolt says individual Pro plans receive up to 50 times more Forge usage at no extra cost through October 14. That is a specific promotional change in the announcement. The source does not explain how usage is measured.

This means readers should treat the offer as time-bound. It also means the exact value depends on the plan and the company's own usage rules. The source does not provide any further pricing details.

Data sharing and privacy choices

Bolt says users who opt in may share anonymized build sessions. The company says this helps Arcee AI train open-weight models. Standard and Max sessions remain private.

This is the clearest governance point in the source. The announcement separates opt-in sharing from private sessions. It also limits the claim to anonymized build sessions, which is important for how users think about data handling.

Because the source is a company announcement, readers should not treat it as an independent privacy audit. The announcement tells us what Bolt says, but not how the process is verified. The source does not provide more detail on retention, access, or deletion.

Reported model score

Bolt also reports a 92.2 score for Forge's open models on its own Bolt Build Index. The company says its top paid model scores 101 on the same index. This comparison appears in the announcement itself.

The source is explicit that this is a company-reported product announcement. It is not an independent benchmark. That distinction matters because the score should be read as Bolt's own framing of performance.

The numbers may still help readers understand how Bolt positions Forge. But they do not establish broader market performance. The source does not include third-party validation or a methodology breakdown.

How to read the announcement

The safest reading is simple. Bolt is introducing a research-preview agent, expanding Pro usage, and offering an opt-in data-sharing path. It is also presenting an internal score comparison for its open models.

Each of those points is useful, but each has limits. The feature is new in the source. The usage boost is temporary. The score is self-reported. The privacy choice is opt-in.

Operational considerations

Users evaluating Forge should focus on three questions. First, does the research-preview label fit their tolerance for early-stage features? Second, does the opt-in sharing model match their data preferences? Third, does the reported score matter for their own workflow?

These are practical questions because the announcement gives only partial detail. The source does not describe reliability, support, or rollout scope beyond the stated plan changes. So any adoption decision should stay close to the announcement's actual terms.

Morocco relevance

The source reports no Morocco-specific availability, market detail, or local program. For readers in Morocco, the global lesson is to check whether a product announcement includes region, plan, and privacy terms before assuming access.

Bottom line

Bolt Forge is a research-preview agent inside Bolt.new that uses open-source AI models. The announcement highlights extra Pro usage, optional anonymized sharing, and a company-reported score comparison.

The main takeaway is that the feature is still framed by Bolt's own announcement. Readers should separate product claims, opt-in data sharing, and internal benchmark numbers. That keeps the evaluation grounded in the source.