Factory Launches Droid 1.0, a Coding Agent That Runs Its Tools From a Script

Factory shipped Droid 1.0 on 29 October, calling it the biggest update yet to its coding agent. The headline change is that the model no longer calls tools one at a time: by default it now writes a small program that calls them for it. Our take: if you already pay for a coding agent and want to switch freely between models, Droid 1.0 is worth a trial run. Just keep in mind that every performance figure comes from Factory itself, and there is no free plan to try it on.

What changed: tools called from a script

In a typical agent loop, the model asks for a tool, the full output lands in its context window, and then it asks for the next one. Droid 1.0 makes what Factory calls Script mode the default. The model writes JavaScript that calls the tools, and that script runs in a sandbox with no file, network or system access of its own. According to Factory’s announcement, one script can make hundreds of tool calls in a single model request, independent calls run concurrently, and only the script’s output goes back into the context.

Factory’s examples show where this helps: reading hundreds of files to find the handful that mention a topic, triaging dozens of Linear tickets, or running a database migration in stages. MarkTechPost, which covered the release on 31 October, framed it the same way: a model-independent agent harness built around Script mode.

Factory’s numbers, and their limits

Factory says Script mode cuts Droid’s cost by about 15% on Claude Opus 5.5 compared with direct tool calls. It also says it trimmed the base prompt sent with every request by 37%, and that a typical request now reads 98% of its input from the provider’s cache. The app itself is quicker to open, and parallel sessions now share one background process, adding roughly 20 MB each instead of about 350 MB.

The company also published a head-to-head it ran between 23 and 28 October. It claims Droid beat Claude Code, Cursor, Codex, OpenCode and Devin on the same models at up to 24% lower cost. To its credit, Factory also reports that Droid was about 10% slower than Codex when both ran GPT-6.1 Sol. Still, this is a vendor testing its own product. Until someone independent repeats it, treat the comparison as marketing.

Who it is for, and what it costs

Droid is aimed at developers who don’t want to be tied to a single model provider. It works with closed and open-weight models; the announcement names Claude Opus 5.5, GPT-6.1 Sol, GLM 5.3 and Kimi K3. You can pick the model yourself or let Factory Router choose, and Factory says the gains also apply when you bring your own API keys. Droid ships as a desktop app for macOS and Windows, as a terminal UI, and as the Droid SDK for Node developers who want to build on it.

Prices on Factory’s pricing page are in US dollars:

• Pro: $20 a month, covering the desktop app, CLI, SDK and background agents.
• Plus: $100 a month, about five times Pro’s usage, plus Factory-managed cloud computers.
• Max: $200 a month, about ten times Pro’s usage and early access to features.
• Teams: $60 a month per team plus $40 a month per seat, up to 10 seats.
• Business and Enterprise: contact sales; adds SSO, audit logs and zero data retention.

There is no free plan, and the limits are described as “rolling” without published numbers, so you can’t tell in advance how far the $20 plan will take you.

What about Arabic?

Neither the announcement nor the pricing page mentions Arabic, regional pricing or availability in Arab countries. How well Droid handles Arabic prompts will most likely depend on the model you choose rather than on the harness, but that is our inference, not something Factory has said. If you build for Arabic-speaking users, run a quick test before committing to a monthly plan. In a real repository, ask Droid in Arabic to explain a module, then have it add Arabic interface strings to a page marked dir=”rtl”. Check three things: whether its replies stay in Arabic, whether its code comments come out in Arabic or English, and whether the right-to-left layout survives.

What to do now

If your daily work involves sweeping changes across many files, try Droid 1.0 on a side project first and compare its bill with your current tool after a week of real use, not after reading the charts. Our Factory listing has the details and links. Remember that the sandbox only applies to scripts: Droid itself still edits your repository directly. Work on a separate branch and review every change before you merge. Just two days ago, the backdoored Claude Code extension in Cursor’s marketplace showed how quickly a coding agent on your machine can become a liability.