Open-model agent
Forge is an agent in Bolt.new's agent picker rather than a single model. Users can switch between Forge, Standard, and Max while working in the app builder.
Bolt Forge
Bolt Forge is an open-model AI agent in Bolt.new for building and iterating on apps with GLM 5.3 Flash and GLM 5.3, designed for experimentation.
What is Bolt Forge?
Bolt Forge is an open-model AI agent within Bolt.new, the browser-based AI app builder. It lets users build with open models selected through the Forge agent rather than relying only on Bolt's Standard or Max agents. The initial lineup includes GLM 5.3 Flash, GLM 5.3, and experimental Kimi K3 and DeepSeek v4 Pro options.
The product is designed to give individual builders more room for drafting, testing, rebuilding, and experimenting. Individual Pro plans receive up to 50 times more Forge usage at no extra cost during the launch period through October 14, 2026. In exchange, users who opt in allow de-identified build sessions to be used in datasets licensed to AI developers; Bolt says Teams and Enterprise workspaces are excluded from this program.
Key features
Selectable model lineup
As of September 2026, the lineup includes GLM 5.3 Flash and GLM 5.3, with Kimi K3 and DeepSeek v4 Pro available as experimental options.
Launch-period usage allocation
Individual Pro plans receive up to 50 times more Forge usage at no extra cost through October 14, 2026. The allowance is separate from Standard and Max usage and uses one monthly bar with no daily limit.
Reserved compute and browser runtime
Bolt says Forge runs on reserved hardware and uses WebContainers in the browser, helping support its usage model without renting a server for every build.
Consent and de-identification
Participation in training is opt-in. Shared prompts, code, project files and configuration, tool calls, and edit histories are de-identified, with secrets and personal information stripped before the data enters licensed datasets.
Predictable usage boundary
Forge usage stops at 100% of the monthly bar and switches back to Standard instead of creating an overage charge. Forge is described as a research preview beginning September 14, 2026, with the special 50-times allocation lasting through October 14, 2026.
Practical use cases
Early-stage product exploration
Use the larger Forge allocation to explore an idea, produce rough drafts, and test alternate approaches before committing premium usage to more complex work in Standard or Max.
Rapid prototype iteration
Build, test, tear down, and rebuild a prototype in Bolt.new when the project needs repeated iteration rather than a single polished generation.
Open-model experimentation
Try open models inside a real Bolt.new project and compare their behavior, using the default GLM option for efficiency and experimental models when the task warrants them.
Consent-based model improvement
Contribute opted-in prompts, code, project configuration, tool calls, and edit histories to de-identified datasets while using Forge for individual development work.
Pros and Cons
Pros
- Provides a separate open-model agent inside Bolt.new, with multiple model choices.
- Offers substantially more usage for eligible individual Pro users during the launch period without an additional Forge charge.
- Uses a single monthly usage bar with no daily limit and switches to Standard at the limit instead of charging an overage.
- Makes training participation opt-in and describes de-identification measures before shared sessions enter licensed datasets.
Cons
- Forge models are described as experimental, so Bolt recommends duplicating a project before moving a serious build into Forge.
- Forge cannot accept PDF uploads yet, and some model choices consume the usage allowance faster than others.
- The 50-times usage allocation is a launch-period offer for individual Pro plans through October 14, 2026; Teams and Enterprise workspaces are excluded from the research preview.
FAQ
How do I use Bolt Forge?
In Bolt.new, select Forge from the agent picker and review its one-tap consent screen before building. Forge usage is drawn from a single monthly bar with no daily limit; when the bar reaches 100%, Bolt switches the user back to Standard rather than charging an overage.
How is Bolt Forge available and how does the usage allowance work?
Forge is included for individual Pro plans at no extra cost through October 14, 2026, with up to 50 times more usage than Standard mode depending on the selected Forge model. Bolt Lite is a separate $9-per-month plan powered by Forge, according to the related Bolt Lite page.
Which models does Forge use?
Forge runs open models including GLM 5.3 Flash and GLM 5.3, with Kimi K3 and DeepSeek v4 Pro listed as experimental options as of September 2026. The model lineup can change as new open models are added.
What data does Bolt Forge collect?
Forge sessions are shared for training only after opt-in. Bolt says it de-identifies sessions, strips secrets and personal information, and validates the pipeline against seeded test data. Switching back to Standard or Max stops new Forge collection. Teams and Enterprise workspaces are excluded from Forge and its training and dataset licensing, and sessions from the EEA, UK, and Switzerland are not used for training or datasets.
What are the current limitations of Forge?
Bolt describes Forge models as experimental inside Bolt and recommends duplicating a project before moving a serious build into Forge. It also says Forge cannot accept PDF uploads yet, and that Kimi K3 and DeepSeek v4 Pro use the allowance faster than the GLM models.
Quick Facts
- Category
- AI app-building and developer tool
- Platform
- Bolt.new, with projects running in browser-based WebContainers
- Agent type
- Open-model agent, not a single model
- Initial model lineup
- GLM 5.3 Flash, GLM 5.3, plus experimental Kimi K3 and DeepSeek v4 Pro
- Eligible launch access
- Individual Pro plans; Forge is excluded from Teams and Enterprise research-preview access
- Research preview
- Begins September 14, 2026; the special up-to-50-times Pro allocation runs through October 14, 2026
Bolt Forge Alternatives
CreateOS Sandbox
CreateOS Sandbox is an isolated compute environment for running code and agent workloads in Firecracker micro-VMs with private networking and SDK, CLI, or MCP control.
GitBot
GitBot is a local browser workspace for reusable coding bots powered by Claude Code, Codex, or OpenCode, with persistent threads and configurable permissions.
ByteAsk
ByteAsk is a terminal-first AI coding agent for C and C++ that edits repos and verifies changes with compilers, debuggers, sanitizers, and tests.
MeetStream
MeetStream is a meeting bot API for Zoom, Google Meet, Microsoft Teams, and Webex. Record, stream, and analyze meetings with usage-based pricing and $5 free credit.
Codex Plugins
Codex Plugins bundle reusable skills, app integrations, and MCP servers into workflows you can install in the Codex app or use from Codex CLI.
hob
hob is an independent workspace for coding agents, with local control over sessions, terminals, history, routing, and follow-up work.