Instagram Agent — AI Configuration

The AI tab (located at /instagram/agent) configures the intelligence of your Instagram Agent (InstagramAgent entity). This is where you connect your Instagram profile to the agent's brain, attach grounding data, pick your preferred LLM, and provide the core prompt instructions governing how the assistant communicates.

1. Instagram Account Linkage

At the top of the AI configuration screen, the Instagram Account card displays the connected profile details (@username, profile photo) along with its current connection status:

• Status Indicator: Shows whether the account is Connected to this agent or Integrated · not linked.

• Connect / Disconnect Toggle:

  • Connect: Binds the account's webhook events to this InstagramAgent instance (account.agent = agent.uuid).
  • Disconnect: Unlinks the account so the agent temporarily stops answering DMs and comments. Unlinking does not delete the account or revoke Meta authorization.

• Pre-requisite: If no account appears, connect one from the Integrations page first.

2. Multi-Knowledge Base Grounding

The Instagram Agent natively supports grounding across multiple Knowledge Bases simultaneously:

• Multi-Select Dropdown: Select one, two, or more Knowledge Bases created in your workspace (e.g., Product FAQs, Shipping Policies, and Pricing Guides).
• Embeddings Retrieval (RAG): When an Instagram user sends a message, Kipps.AI performs semantic vector search across all attached Knowledge Bases to pull the most relevant context before querying the LLM.
• Training: Make sure to click Train on your Knowledge Bases whenever you add new documents or update URLs.

Tip: Combining a broad FAQ knowledge base with a dynamic inventory or product catalog knowledge base allows the agent to handle diverse customer questions without hallucinating.

3. LLM Provider, Model & BYOK

Customize the inference engine powering your agent's reasoning:

Provider & Model Selection

• Provider Filtering: Choose from top providers including Google, OpenAI, Anthropic, or Groq.
• Model Selection: The model dropdown dynamically filters to display models for the selected provider (e.g., gemini-2.5-flash, gpt-4o, claude-3-5-sonnet).
• Default Model: If no model is explicitly selected, the agent defaults to gemini-2.5-flash for high-speed, cost-effective inference.

Bring Your Own Key (BYOK)

• Under User API Key, select an API key configured in your Provider API Key settings.
• If no custom key is selected, requests use your organization's included platform credit allocation.

4. Temperature Slider

Controls the trade-off between strict predictability and conversational creativity:

5. Unified System Instructions

Unlike legacy configurations that split prompts across multiple disparate screens, the Instagram Agent uses one comprehensive instructions prompt that governs all interactions:

Default Prompt Structure

<role>
You are a friendly, helpful assistant that replies to Instagram DMs and comments on behalf of the brand. Keep replies short, warm, and on-brand — Instagram is casual. Use the customer's name when you know it, and add a light emoji when it fits.
</role>

<guidelines>
- Answer only from the training data / knowledge base. If you don't know, say so honestly and offer to connect them to a human.
- Never invent prices, availability, order status, or policies — use a function/tool if one is available, otherwise ask.
- Keep it to 1-3 short sentences. No walls of text.
- Be polite even if the customer is rude. End on a positive note.
</guidelines>

Prompt Engineering Best Practices for Instagram

1. Keep Responses Concise: Instagram users prefer rapid, 1 to 3 sentence replies over large blocks of text.
2. Handle Boundaries Gracefully: Direct the agent to offer a human handoff when customer queries exceed its knowledge scope.
3. Encourage Action: When appropriate, instruct the agent to direct users to specific links, booking slots, or pre-built templates.

6. Saving Changes

Once your Knowledge Bases, model, temperature, and instructions are set:

1. Click the Save button in the top right or bottom of the configuration card.
2. The agent updates instantly without requiring a redeployment or server restart.

Frequently Asked Questions

Can I attach multiple Knowledge Bases to my Instagram Agent? Yes. The agent supports Multi-KB. You can select multiple knowledge bases simultaneously (for example, combining a general FAQ knowledge base with a product catalog or pricing sheet). The agent searches across all selected KBs before generating replies.

What happens if no Knowledge Base is selected? The agent will rely exclusively on its base LLM model weights and your custom system instructions. For brand-accurate responses and preventing hallucinations on pricing or policies, attaching at least one trained Knowledge Base is strongly recommended.

Can I use my own OpenAI, Anthropic, or Gemini API key (BYOK)? Yes. Under the User API Key selector, choose any pre-configured key from your Provider API Keys settings. If left unselected, requests utilize your organization's included platform credit allocation.

What is the difference between connecting the account here versus in Integrations? The Integrations page establishes the OAuth authorization between Meta and your organization. The AI tab's "Instagram account" card links that authenticated account to this specific InstagramAgent entity so the agent begins answering its webhook events.

How often should I retrain my Knowledge Bases? Whenever you add new documents, edit text, or update scraped websites in the Knowledge Base area, click Train on that KB so newly created embeddings are available to your Instagram Agent.