Factory describes Droid 1.0 with programmatic tool calling

Factory defines an agent harness as the prompts, tools, tool-result handling and execution loop surrounding a model.[1] It presents Droid 1.0 as a new engine for its enterprise agent, intended for complex projects as well as everyday development.[1] Users can choose a model themselves or let Droid select an efficient mix using Factory Router.[1] For teams evaluating the system, keep the harness and the underlying model choice distinct: the announcement describes changes to how the agent works around a model, not a single required model.

With programmatic tool calling, an agent can write a tool-calling workflow and execute it in one invocation.[1] The model writes JavaScript that runs in a sandbox without its own file, network or system access.[1] A Script can make hundreds of tool calls within a single model request, and Factory says ordinary tool outputs remain inside the program as variables.[1] Factory argues that running many calls through one program reduces model requests, which can make complex tasks faster and cheaper.[1] Teams should review how that sandbox boundary and tool workflow fit their own operating requirements.

Factory says programmatic tool calling improves reliability because checks and loops run deterministically in code.[1] In an example fixing a failing unit test, Script mode used five model calls and finished in 41.6 seconds, compared with 10 calls and 56.8 seconds for direct tool calls.[1] Factory also reports that Script mode cuts Droid's cost by 15% on Claude Opus 5.5.[1] Its listed scores are Droid 65.2%, Cursor 62.1% and OpenCode 62.1%.[1] Factory says its Claude Opus 5.5 score was 3.0 to 7.6 points higher than those of Cursor, OpenCode, Devin and Claude Code, at up to 24% lower cost.[1] These are Factory's reported results, so teams should evaluate them against their own tasks.

Droid 1.0 is available as a desktop app for macOS and Windows, through a terminal interface and through a TypeScript SDK.[2] Factory lists its Pro plan at $20 per month, including desktop, CLI and SDK access.[3] Neural Brew reports that Droid's initial prompt shrank from 20,887 to 13,055 input tokens between versions 0.170.0 and 0.250.0, a 37% reduction.[2] It attributes much of that change to sending two tool definitions up front instead of 11.[2] Neural Brew said it had not independently verified the claims.[2] Readers should treat the performance and cost figures as vendor-reported evidence rather than an independent assessment.