AI builder setup

The AI builder lets you describe what you want and Swebsy generates or edits page sections for you. Two provider types are supported:

Provider	Best for
Anthropic (Claude)	Highest quality results; requires an Anthropic account and API key
OpenAI-compatible / Local	Privacy-first or offline use; works with Ollama, LM Studio, Groq, Together, and any server that speaks the OpenAI /v1/chat/completions API

---

Option A — Anthropic Claude

Before you start

• An Anthropic Console account - An API key with access to Claude models

1. Click Settings in the topbar.
2. Click the AI tab.
3. Set AI provider to Anthropic (Claude).
4. Paste your Anthropic API key into the API key field.
5. Select the Claude model you want to use (default: Claude Sonnet).

6. Optionally set a Monthly budget cap to receive a warning when the session cost exceeds it.

Expected result

The AI chat panel becomes active. You can now type prompts to generate or edit content.

Available Claude models

Model	Speed	Quality	Recommended for
Claude Haiku 4.5	Fastest	Good	Quick edits and small changes
Claude Sonnet 4.6	Balanced	High	General use (default)
Claude Opus 4.7	Slowest	Highest	Full page generation

You can also type any Anthropic model ID directly by selecting Custom model ID… from the model dropdown.

---

Option B — OpenAI-compatible / Local

This option works with any server that implements the OpenAI /v1/chat/completions API — including local apps running on your machine.

Before you start

• A running OpenAI-compatible server (see examples below) - The model name as the server expects it (e.g. llama3, mistral)

Popular local servers

Server	Default base URL	Notes
Ollama	http://localhost:11434	Run ollama serve then ollama pull llama3
LM Studio	http://localhost:1234	Enable the local server from the app

1. Start your local AI server (e.g. run ollama serve).
2. Click Settings in the topbar, then the AI tab.

3. Set AI provider to OpenAI-compatible / Local (Ollama, LM Studio…).

4. Enter the Base URL of your server (e.g. http://localhost:11434).

5. Enter the model name exactly as the server expects it (e.g. llama3.2).

6. If your server requires authentication, paste an API key into the API key field. Local servers (Ollama, LM Studio) do not require one — leave it blank.

7. Click Test connection to verify the server is reachable.

Expected result

“Connected” appears next to the Test connection button. The AI chat panel is ready to use.

Intent routing on local providers

When using a local server, Swebsy skips the AI-powered intent classification step and uses keyword-based routing instead. This is faster and avoids an extra API call. For most prompts the results are the same; very ambiguous prompts may default to “edit section” when another operation would be a better fit.

---

Add an Unsplash key (optional)

When an Unsplash access key is configured, the AI replaces placeholder image URLs with real photos fetched by keyword from Unsplash.

1. Create a free app at unsplash.com/developers.
2. Copy the Access Key.
3. Paste it into the Unsplash access key field in Settings → AI.

---

Use the AI builder

After saving your settings, open the AI panel from the editor and describe the change you want. The AI builder can help with common authoring tasks:

Task	What to ask for
Create a page	”Create a landing page for a photography studio.”
Add a section	”Add a pricing section below the hero.”
Edit selected content	”Make this section clearer and more direct.”
Update site settings	”Set the site title to Acme Studio and improve the meta text.”
Match an existing style	”Add a testimonials section that matches this page.”

For best results, select the page section you want to change before asking for an edit. Use short, specific prompts and review the generated content before publishing.

Per-operation model overrides

In the Model overrides section you can assign different models to different operations:

Operation	When it fires
create_page	Generate an entire new page
add_section	Add a new section to an existing page
edit_section	Edit a selected section
update_settings	Change site settings via chat

Use this to run a more powerful model only for full page generation, while keeping a faster model for quick edits.

Model overrides are most useful with Anthropic where multiple Claude tiers exist. With a local server, just enter the model name you want in the main Model field.

Budget expectations

Anthropic: Usage is billed by Anthropic against your account. Set a monthly budget cap in Settings → AI to get a session warning before costs add up.

Local / OpenAI-compatible: Running models locally is free of Anthropic charges. Cloud providers on the OpenAI-compatible path (Groq, Together, etc.) bill through their own accounts — check their pricing directly.

Privacy

• Your API key is stored only in your browser — it never leaves your device, and it is not included in project backups or exports.
• Local servers (Ollama, LM Studio) never send your data to a third-party — everything stays on your machine.
• Enable Sanitize PII to have Swebsy redact emails and phone numbers before sending prompts to any AI provider.

Common issues

“Invalid API key” error (Anthropic) — check that you copied the full key from the Anthropic Console without any extra spaces.

“Cannot reach server” error (local) — make sure the server is running before testing. For Ollama: ollama serve. For LM Studio: enable the local server from the app’s settings.

AI panel doesn’t appear — ensure your settings were saved, then refresh the page.

Responses are slow — for Anthropic, try switching from Opus to Sonnet or Haiku. For local servers, a smaller or quantized model will be faster.

Model not found (local) — the model name in Swebsy must match exactly what the server expects. For Ollama, run ollama list to see available model names.