Getting started · 2026-09-24
From a request
to project files.
Donatello Agent is a free AI coding agent for Windows. Describe a website, app, game or repair: it can edit project files, run development tools and help you test the result on your PC.
1. Choose your Windows edition
Microsoft Store edition is recommended for new installations and is updated by Microsoft Store. The direct EXE edition uses its own installer and in-app updater. They have separate profiles and chat histories; installing one does not automatically migrate the other. Keep the edition containing your existing work.
Compare Store and direct EXE downloads.
2. Start with a small, testable project
Create a new project, or open a folder you intend the agent to work on. Back up existing work first. Ask for an outcome and acceptance checks: “Build a static restaurant website with a mobile menu; make clear that its contact form is only a demo. Use no paid services. Test the links.” Review permission requests and stop the task if its approach is not what you intended.
3. Understand the free routes
The download and available free AI routes require no subscription or positive Donatello credit balance. Optional Donatello Local runs on your computer without per-request Donatello credits or a provider usage quota after its model is installed.
You do not need to supply an API key to start with an available built-in free route, or to run the installed local model. Optional provider connections may need your own key or authorization and follow that provider's terms and quotas. Credit-based Donatello services require a linked account.
Automatic mode prioritizes eligible online free routes. If they fail, an installed compatible local model can be tried for text-only tasks. The current interface does not provide a local-only chat selector. If no suitable route can run, the task can pause or fail; a credit-based continuation requires permission.
Donatello AI and selected online media services use optional credits. Review and approve the relevant quote or credit-continuation request before proceeding. Earned credits can be used too; buying credits is optional. Free mode is not a promise that every tool or external service is free.
The model shown in the app is the current choice for that task, not a guarantee of permanent availability. No static list here promises continued access to a particular online model.
4. Optional Donatello Local
Open Settings, find the Donatello Local card and choose the free local-AI activation action. The initial model download needs Internet and sufficient storage. Wait until verification finishes and the model is ready. This enables a local fallback; it is not a switch that forces every future task to stay offline.
Donatello Local uses IBM Granite 4.1 3B Instruct, Q4_K_M, with a 16,384-token context. Activation downloads a model of 2,099,501,664 bytes (about 2.10 GB) and checks its SHA-256. The bundled runtime supports CPU and compatible GPU acceleration. Local inference can work offline after installation; online search, dependency downloads and cloud media still need Internet. This text-only model does not generate images, voice or video by itself.
Windows 10 or 11, x64. The local model needs about 2.10 GB of disk space in addition to the app, projects and runtime memory. RAM and performance depend on context, other applications and project tools; no measured minimum RAM or universal speed is promised here. A dedicated GPU is not required for cloud AI; local CPU execution may be slower. Godot, Expo or other project-specific tools may need separate installation.
Project files are saved on your PC. Online AI services may receive prompts and relevant file or tool context. Local inference runs on your computer, but automatic mode is not a local-only privacy guarantee: it prioritizes available online routes. Review permissions and do not include secrets in prompts. Optional usage telemetry is disabled by default.
5. Test before publishing
Generated files and a local preview are not a live website. Review the code, dependencies, security and licenses, then deploy to a compatible hosting provider. Hosting accounts, domains, credentials and any hosting costs are separate; Donatello does not promise free hosting for every project.
Check the actual result: open the site, exercise controls, run the project's tests and review third-party dependencies. An agent's “done” message or a generated screenshot is not proof that every workflow works. Read the methodology and inspect available case evidence.
Practical questions
Do I need an API key?
You do not need to supply an API key to start with an available built-in free route, or to run the installed local model. Optional provider connections may need your own key or authorization and follow that provider's terms and quotas. Credit-based Donatello services require a linked account.
Is there free local AI without a usage quota?
Donatello Local uses IBM Granite 4.1 3B Instruct, Q4_K_M, with a 16,384-token context. Activation downloads a model of 2,099,501,664 bytes (about 2.10 GB) and checks its SHA-256. The bundled runtime supports CPU and compatible GPU acceleration. Local inference can work offline after installation; online search, dependency downloads and cloud media still need Internet. This text-only model does not generate images, voice or video by itself. Automatic mode prioritizes eligible online free routes. If they fail, an installed compatible local model can be tried for text-only tasks. The current interface does not provide a local-only chat selector. If no suitable route can run, the task can pause or fail; a credit-based continuation requires permission.
What happens when online free routes are unavailable?
Unlimited means no fixed lifetime message allowance, not guaranteed uninterrupted service, one particular premium model or unlimited computing power. Online free routes can be rate-limited or unavailable. Local work is limited by your PC, model capabilities and task complexity. Automatic mode prioritizes eligible online free routes. If they fail, an installed compatible local model can be tried for text-only tasks. The current interface does not provide a local-only chat selector. If no suitable route can run, the task can pause or fail; a credit-based continuation requires permission.
What computer do I need?
Windows 10 or 11, x64. The local model needs about 2.10 GB of disk space in addition to the app, projects and runtime memory. RAM and performance depend on context, other applications and project tools; no measured minimum RAM or universal speed is promised here. A dedicated GPU is not required for cloud AI; local CPU execution may be slower. Godot, Expo or other project-specific tools may need separate installation.
How do I publish my project?
Generated files and a local preview are not a live website. Review the code, dependencies, security and licenses, then deploy to a compatible hosting provider. Hosting accounts, domains, credentials and any hosting costs are separate; Donatello does not promise free hosting for every project.
Can I use my projects commercially?
Commercial projects are possible, subject to Donatello's Terms and the licenses of models, dependencies and assets. The local model is distributed under Apache-2.0. This does not guarantee exclusive copyright, non-infringement or permission to use third-party names, images or code.
Can it use Blender, Godot, Unity or Unreal Engine?
Beyond websites, Agent can help create games, 3D scenes, local videos, presentations and utilities through project files, scripts and installed tools. Godot has a dedicated validation and launch adapter; Blender is detected as a local tool. Unity and Unreal Engine workflows use their configured tools and public scripting or command-line interfaces, not a dedicated Donatello integration. Tools must be installed, compatible and licensed where required. Free Agent access does not include third-party licenses, assets, hosting or cloud services. No benchmark here establishes superiority over other agents.