Getting started

Getting started

The goal of the first session: connect a chat provider, index your products, open the chat widget on your storefront, and have one conversation.

1. Connect a chat provider

OAssist -> Models. Seven provider slots are pre-seeded inactive:

• OpenAI (default model gpt-4o-mini)
• Anthropic (default model claude-haiku-4-5)
• Google Gemini (default model gemini-2.0-flash)
• Azure OpenAI
• Groq (default model llama-3.3-70b-versatile)
• OpenRouter (default model openai/gpt-4o-mini)
• OpenAI Embeddings (default model text-embedding-3-small)

Click a provider row, paste your API key, hit Test Connection. The key is encrypted at rest via OAssist_Crypto and stored in oas_providers.api_key_encrypted. Mark one chat provider as Default for the assistant to use it.

OAssist is BYOK by design - no inference markup, no caps. You pay your provider directly.

2. Index your knowledge

OAssist -> Knowledge. Index sources for retrieval-augmented generation (RAG):

• Products: indexes WooCommerce product titles, descriptions, attributes.
• Pages and posts: indexes your CMS content (FAQ pages, shipping policies).
• Custom URLs: indexes external pages you specify.
• Files: upload .txt, .md, .pdf for indexing.

Click Index Now on the source. OAssist chunks the content, generates embeddings via your configured embedding provider, and stores them in oas_chunks.

3. Configure persona

OAssist -> Persona. Set:

• Bot name (shown in the widget header).
• Tone (helpful, professional, casual, witty).
• Welcome message (first message visitor sees).
• Out-of-scope behaviour (refuse politely, escalate to human, suggest contact form).

4. Place the chat widget

Two options:

• Floating launcher: enable under Channels -> Site. The widget appears on every storefront page.
• Inline block/shortcode: drop the OAssist Chat block, or use [oassist_chat height="520px"] in a page.

5. Have a conversation

Open your store in an incognito window. Ask: "Do you have wireless headphones under $50?" The assistant should:

1. Match the intent.
2. Call the search_products tool with { query: "wireless headphones", max_price: 50 }.
3. Render the matching products with links.
4. Open a row in oas_conversations and message rows in oas_messages.

OAssist -> Inbox shows every conversation in real time with full transcript, tool calls, and cost per turn.