Agent - Anything

Documentation Index

Fetch the complete documentation index at: /docs/llms.txt

Use this file to discover all available pages before exploring further.

The core of Anything is its agent. You describe what you want, the agent writes the code. It reasons about your request, explores your project, decides what to change, and writes or updates code.

See the Prompting Guide for tips on writing better prompts.

Plans

Your plan determines which models and capabilities the agent uses.

• Pro unlocks stronger models, deeper reasoning, more credits, and more context.
• Max gives you the most capable models, plus autonomous features like browser testing, test running, and background tasks.

See pricing.

Code generation

It generates React for web apps and React Native for mobile apps. You don’t need to know either. Describe what you want in plain language. Tell it what your app should do, not how to implement it. “Add a settings page where users can update their profile” beats “create a React component with useState hooks for form fields.”

Modes

Switch modes in the mode selector at the bottom of the chat. New projects start in Thinking so you’re on the most efficient credit mode by default, with control over which mode runs each request.

• Thinking is the default and is best for most requests. More accurate, but slightly slower.
• Fast is good for simple tasks. Faster and uses fewer credits, but less accurate.
• Discussion lets you talk to the agent without it touching your code. Use it to ask questions, get feedback, or brainstorm. Nothing you say here will change your app.
• Plan explores your codebase and creates a step-by-step implementation plan before making any changes. Good for complex features where you want to review the approach first.
• Max is autonomous. Give it a goal and it builds, tests, and fixes on its own.

Thinking handles most requests well, so you can leave it set there. Pick Fast for simple changes to save credits, and Plan when you want to see the approach before the agent starts building.

Some accounts also have an Auto mode that picks the approach for each task. We’re rolling out Thinking as the default instead, so you control which mode runs rather than letting Auto decide. If you don’t see Auto in your selector, that’s expected.

Model switcher

The model selector (next to the mode selector) lets you override the agent’s model choice. By default, the agent uses multiple models and picks the best one for each step. Models are available from Anthropic (Claude series), OpenAI (GPT + Codex series), Google (Gemini series), and Moonshot (Kimi series). Some are more capable, others are faster or use fewer credits.

New models are added regularly. Some may be marked as beta.

Not sure which model to pick? Leave it on the default. The agent picks the best model for each step automatically.

Thinking

Work gets broken into steps. Expand each step to see the reasoning: why it made certain decisions, what it considered, how it approached the problem. Useful when something doesn’t work as expected.

Capabilities

It does more than write code:

• Web search. Give it a URL or ask it to search for reference material.
• Integrations: press / to add AI services and other tools. The agent wires them up.
• Image generation. Creates images and assets for your app (icons, illustrations, backgrounds).
• Error fixing tries to fix errors on its own. You can also say “fix me” or paste error messages into chat.
• Version history. Finds and suggests previous versions to restore. Shows an interactive card in chat with a one-click Restore button. See Version history.
• Codebase exploration. Searches your project files to understand context before making changes. You’ll see a live activity indicator showing what it’s looking at.
• Runtime logs. Reads your app’s logs to diagnose errors without you having to copy-paste them.
• Images. Paste or drag screenshots, mockups, or design references anywhere on the page. The agent uses them to match styles and layouts.
• File attachments. Attach PDFs, spreadsheets, docs, and other files to your chat message. The agent reads them and uses the content to inform your app. Supported formats: PDF, CSV, XLSX, DOCX, TXT, HTML, JSON, ODT, RTF, EPUB.
• Figma import. Paste a Figma link and the agent builds from your design.

In Max mode, the agent also gets:

• Browser testing. Opens your app in a real browser, clicks through it, finds issues.
• Mobile testing. Runs your app on a simulated iOS device to verify native interactions.
• Test running: runs your backend logic to verify it works.
• Background tasks let it work on larger goals while you do other things.

Threads

Press New chat in the left sidebar to start a new thread. Each thread has its own context.

1. Start your app in one thread
2. Start a new thread for each feature
3. Work across threads in parallel

This keeps the agent focused and avoids context overload.

Fixing errors

Most errors happen because the agent doesn’t have enough context.

• Be specific. What’s wrong, what you expected, what happened instead.
• Paste the error: copy error messages from the app, logs, or your browser into chat.
• Let it try. It fixes errors on its own by default. If it misses, give it more detail.

See the Prompting Guide for more.

Credits

Credits are used per task. Cost depends on complexity, which models run, and how much context is involved. Higher-tier plans include more credits and stronger models. Save credits: keep prompts focused, use new threads to reduce context, use Fast mode for simple changes.