Droid 1.0 calls tools through programs

Factory’s announcement describes an agent harness: the prompts, tools and execution loop surrounding a model. In Droid 1.0, programmatic tool calling becomes the default. The agent writes JavaScript that invokes tools. Raw files, search results and logs remain program variables; printed output returns to the model. The sandbox itself has no file, network or system access. Independent tool calls can run concurrently, while later calls can use earlier results.

The CLI quickstart starts with a terminal inside a code project; Git is recommended. After installation, run droid there and sign in through the browser if prompted. First ask it to explain the architecture, entry points and tests. Then request a small change. The documented workflow inspects the implementation, proposes a plan when needed, shows the exact diff and waits for approval before editing. Validators run when you ask it to finish. Review the diff and run relevant tests before accepting the change.

MarkTechPost’s coverage also discusses performance, explicitly identifying the benchmark figures as Factory’s own results. Its comparison points readers to the provider’s methodology. That distinction matters: this is coverage of company measurements, rather than an independent evaluation of the product.