AI and API keys

Every AI and stock content feature in the editor runs on keys you provide yourself. This page covers where to enter them: the AI provider used by the chat assistant, the five stock media providers used by the Media tab, and the optional vector-search knowledge base behind the assistant's design recipes.

Quick overview

1. Open Settings, then AI Settings.
2. Pick a provider tab, paste in your API key, and choose a model.
3. Set an optional system prompt, temperature, and max tokens under General Settings, then Save Settings.
4. Use Test Connection to confirm the key works before relying on it.
5. Switch to the Knowledge Base section, or scroll to Stock Media API Keys, for the other two key sets.

Detailed reference

AI providers

Provider	Notes
Qwen (DashScope)	The default provider
Kimi (Moonshot)
DeepSeek
GLM (Zhipu)
OpenAI
Google Gemini
Anthropic Claude
Custom endpoint	Any OpenAI-compatible chat completions API

Each tab shows a password-masked API key field with a show/hide toggle, and a model picker as a grid of cards. The Custom Endpoint tab replaces the model grid with a Base URL and Model Name field instead, since there is no fixed model list to choose from. See AI chat assistant for which providers support vision and native image generation.

General settings

Shared across every provider.

Setting	Range	Notes
System prompt	Free text	Optional custom instructions, for example steering the assistant toward a particular design style
Temperature	0 to 2	Labeled Precise, Balanced, Creative, or Wild as you move the slider
Max tokens	100 to 32768	Caps how long a single response can be

Save Settings stores the active provider, key, and model; Test Connection sends a small request to confirm the key is valid without waiting for a real chat.

Stock Media API Keys

A separate list of keys that power search inside the Media library, unrelated to the chat assistant's own key.

Provider	Used for	Signup	Notes
Unsplash	Image search	unsplash.com/oauth/applications	Uses an Access Key, not the Secret Key; roughly 50 requests/hour on an unreviewed demo app, much higher once your app is approved for production
Pexels	Image and video search	pexels.com/api	The key is issued instantly after you request it
Pixabay	Image, video, and audio search	pixabay.com/api/docs	One key covers all three media types, with a per-minute rather than per-day limit
Freesound	Audio search	freesound.org/apiv2/apply	Sent as a token, not the raw key value, in the request
Jamendo	Audio search	devportal.jamendo.com	Free for non-commercial use
GIPHY	GIFs and stickers	developers.giphy.com	Also used by the Emoji, GIFs, and stickers panel

Each field has a Show/Hide toggle and inline signup instructions. A Test Keys action checks every configured key at once and reports a pass/fail per provider, including the HTTP status code on a failure, so a bad or expired key is easy to isolate from a provider that is simply down.

Knowledge Base

An optional Qdrant vector database that backs the assistant's semantic search over its built-in design recipes. Without it, the assistant falls back to local keyword matching, which still works but is less precise.

Field	Purpose
Qdrant URL	Your Qdrant Cloud (or self-hosted) endpoint
Qdrant API key	Authenticates against that endpoint
Embedding provider	OpenAI, Qwen, Google, or GLM
Embedding model	Depends on the chosen provider
Embedding API key	Optional; leave empty to reuse your AI chat provider key

Save & Enable turns Qdrant search on, Seed Qdrant pushes the built-in recipe set into it, and Disable Qdrant reverts to local search without losing the connection details.

Skills Manager

The Customize / Skills button at the top of AI Settings opens the skill editor, where you create and assign AI skills, custom instruction profiles triggered with a slash command. See AI skills for how to build and use them.

A skill is structured like a small folder: a main SKILL.md instruction plus an optional set of reference documents, each with its own title and content, that the skill can draw on.

Element	Detail
Built-in skills	Ship with the editor; their name, description, and SKILL.md are read-only, but you can still assign them to providers
My Skills	Anything you create; fully editable, including SKILL.md, references, icon, and description
Provider assignment	A checkbox grid lets you turn a skill on for one or more providers independently, so different providers can run different skills
Slash command	Each skill's name is slugified into a chat command (shown live as you edit the name), for example a skill named "Brand Voice" becomes /brand-voice
Duplicate	Works on both built-in and custom skills; duplicating a built-in skill is the way to start editing it, since the original stays locked

Knowledge Base connection details

The Knowledge Base section (a separate sidebar entry from AI Settings) is where the Qdrant URL, key, and embedding provider from the table above are entered and saved; Save & Enable, Seed Qdrant, and Disable Qdrant all live in that section.

Step by step

Connect an AI provider key end to end

1. Open Settings, then AI Settings.
2. Click a provider tab (for example Qwen, DeepSeek, or Anthropic Claude).
3. Paste your key into the API Key field, using the show/hide toggle to double check it, then pick a model from the card grid (or, on Custom Endpoint, fill in a Base URL and Model Name instead).
4. Optionally set a system prompt, temperature, and max tokens under General Settings.
5. Click Save Settings, then Test Connection to confirm the key actually works before relying on it in a chat.

Set up a stock media key and verify it

1. Scroll to Stock Media API Keys further down AI Settings.
2. Sign up at the provider's site listed above and copy the key it issues.
3. Paste it into the matching field and click Save API Keys.
4. Click Test Keys; each configured provider reports OK or a specific HTTP error inline, so you know immediately if a key needs fixing.

Build a custom AI skill and trigger it

1. Click Customize / Skills at the top of AI Settings.
2. Click New Skill, give it a name and description, and pick an icon.
3. Write your instructions in the SKILL.md editor; add one or more reference documents from the file tree if the skill needs supporting material beyond the main instruction.
4. Check the provider(s) you want it available on in the Provider Assignment grid, then save.
5. In the AI chat, type the skill's slash command (shown in the preview under the editor) to trigger it.

Turn on the Qdrant knowledge base

1. Open Settings, then Knowledge Base.
2. Enter your Qdrant URL and API key, pick an embedding provider and model, and optionally a separate embedding API key (leave it empty to reuse your AI chat provider key).
3. Click Save & Enable, then Seed Qdrant to push the built-in recipe set into it.
4. If you want to go back to local keyword matching, click Disable Qdrant; your connection details are kept, so re-enabling later does not require re-entering them.

Common tasks

Task	Steps
Switch the active provider without losing other keys	Click a different provider tab and save; each provider's key is stored independently
Give one skill to only your fastest provider	Uncheck it for every other provider in that skill's Provider Assignment grid
Start editing a built-in skill	Open it, click Duplicate, then edit the copy in My Skills
Confirm a Custom Endpoint is really working	There is no model list to sanity-check against, so send a real chat message after saving rather than relying only on Test Connection
Free up a knowledge base connection without losing the recipes seeded into it	Use Disable Qdrant instead of clearing the URL and key fields

Troubleshooting

A missing key shows a prompt, not a silent failure

Both stock search panels and the GIF/sticker items panel show an explicit "add your key" message linking back into Settings when a required key is absent, rather than returning empty results.

Test Keys reports per-provider HTTP status

A failed test shows the actual HTTP status code returned by that provider (for example a 401 for a bad key versus a 429 for rate limiting), so you can tell a typo apart from a temporary provider outage.

Built-in skills cannot be edited or deleted in place

Their name, description, and SKILL.md fields are read-only, and there is no delete button. Duplicate a built-in skill first to get an editable, deletable copy in My Skills.

Tips

Keep the chat and stock keys separate

The AI provider key drives conversation and generation; the stock media keys only power search inside the Media tab. Setting one does not configure the other.

Test before you rely on it

Provider APIs change their key formats and rate limits over time. Run Test Connection or Test Keys after pasting a key in, rather than discovering a typo mid-design.

Related

• AI chat assistant for what each provider supports
• AI skills for building and triggering custom skills
• Media library and stock content for what the stock keys unlock
• Editor settings for the rest of the Settings screen