Get AI-Generated Reply Suggestions

When your message input is empty and the last message is from the customer, AI can draft a reply for you. It reads the last 10 messages in the conversation and generates a contextual response that matches your company's tone settings.

How to get a suggestion

1
Open a conversation with a customer message
The AI suggest button only appears when the last message is from the customer and your input field is empty.
2
Click the sparkles icon or press the shortcut
Look for the sparkles icon (✦) in the message toolbar. Or use the keyboard shortcut: Cmd+Shift+S on Mac, Ctrl+Shift+S on Windows/Linux.
3
Review and edit the suggestion
The AI-generated text fills your input field. Read it, adjust anything that doesn't sound right, and send when ready.

Chat input with the AI Suggest button visible in the toolbar, ready to generate a reply suggestion based on the customer's question

Quick Reply matching

When generating a suggestion, AI also checks your Quick Reply library. If one of your existing templates is a good match for the customer's question, it's suggested instead of generating new text. This keeps responses consistent with your pre-approved messaging.

Caching: Suggestions are cached for 5 minutes per conversation. If you open the same conversation twice within that window, you'll get the same suggestion without a new API call. The cache resets when a new message arrives.

Pro Tip: Build a solid Quick Reply library first. AI will match against your templates before generating from scratch, giving you more predictable and on-brand responses.

Success Check: You've generated your first AI suggestion. The sparkles icon appeared, you clicked it, and a contextual reply filled your input field.

Polish & Transform Your Drafts

Already started typing? AI switches to editing mode. The icon changes from sparkles to a wand, and the behavior depends on how much you've written.

Polish Mode

When your draft is ≤100 characters

Light touch improvements. Fixes grammar, spelling, and minor tone issues. Your structure and length stay the same. Think of it as a quick proofread before hitting send.

Transform Mode

When your draft is >100 characters

Full restructure into professional customer support format. Adds a proper greeting, organizes your content into clear sections, and appends your company sign-off. Your rough notes become a polished response.

Using Polish or Transform

1
Type your draft in the message input
Write whatever comes to mind. The AI will handle the polishing or restructuring.
2
Click the wand icon
The wand icon replaces the sparkles when text is present. The tooltip shows "Polish" for short text or "Transform" for longer text.
3
Review the result
The processed text replaces your draft. Check that it captures your intent, then send.

Message input showing a rough draft with the Transform button highlighted in the toolbar, ready to polish the response into a professional message

Pro Tip: Use Transform when you've jotted down key points but don't have time to format them properly. It adds your configured greeting, structures the content, and appends your sign-off automatically.

Success Check: You've used both Polish (short text fix) and Transform (full restructure) on your drafts. The wand icon appeared and the processed text replaced your original input.

Translate Messages with Context

When a customer writes in a different language, a translate button appears next to their message. Translation is context-aware — it uses your company name and conversation history to produce accurate, natural translations across 18 supported languages.

How translation works

Inbound messages

When a customer message is in a different language, a translate button appears. Click it to see the translation inline, below the original text. The original message is always preserved.

Outbound messages

Type your reply in your language, then click the translate button on the send bar. Preview the translation before sending. A flag indicator on the send button shows the target language.

Translation works independently from suggest, polish, and transform. You can generate a suggestion with AI, then translate it before sending — the features compose naturally.

Pro Tip: Translation uses conversation context to handle industry-specific terms and your company name correctly. The more conversation history available, the more accurate the translation.

Success Check: You've translated an inbound customer message and sent a translated outbound reply. The original text was preserved alongside the translation.

Configure AI Tone & Style

Owner Only

AI output follows your company's style settings. Configure these once and every suggestion, polish, and transform will match your brand voice.

Finding AI settings

Click the Replies icon in the sidebar, then select the AI tab (the fourth tab, after Quick Replies, Auto Replies, and Variables).

Available settings

Setting	Options	What it affects
Tone	Professional, Friendly, Casual	Overall voice of AI responses
Formatting	Minimal, Detailed	Structure depth in Transform mode
Greeting	Custom text (max 50 chars)	Opening line in Transform mode (e.g., "Hello,")
Sign-off	Custom text (max 50 chars)	Closing line in Transform mode (e.g., "Best regards,")
Custom Instructions	Free text (max 200 chars)	Extra style guidance for all AI modes

AI Settings tab under Replies showing the tone dropdown set to Professional, formatting toggle, greeting and sign-off text fields, and custom instructions textarea

Pro Tip: Use Custom Instructions for recurring preferences like "Always mention our 24-hour support guarantee" or "Keep replies under 3 sentences." This saves your agents from editing the same things repeatedly.

Success Check: You've configured your AI tone and style settings. All future AI-generated responses will follow these preferences.

Bring Your Own API Key

Owner Only

By default, AI features use a shared API key with standard rate limits. If you need higher throughput, want direct cost control, or have compliance requirements, you can provide your own API key.

Setting up BYOK

1
Get an API key from any OpenAI-compatible provider
Converge works with any provider that exposes an OpenAI-compatible /chat/completions endpoint. Examples: OpenAI, OpenRouter, Cerebras, Groq, Together AI, or a self-hosted model.
2
Open AI Settings
Navigate to Replies → AI tab and scroll to the AI Provider section.
3
Configure your endpoint
Set the Base URL to your provider's API base (e.g., https://api.openai.com/v1 — without /chat/completions). Set the Model to the model identifier your provider uses. Enter your API Key.
4
Save
Your key is encrypted at rest. All AI requests will now use your provider instead of the shared one.

Security: API keys are encrypted with per-company encryption keys before storage. They are never logged, never sent to the frontend, and are only decrypted server-side when making AI requests.

Success Check: Your API key is saved and encrypted. AI requests now use your provider with your own rate limits and billing.

Best Practices & Tips

Always review before sending

AI output is a starting point, not a final answer. Read every suggestion and verify facts, links, and promises before hitting send. Your customers are talking to you, not to an AI.

Build your Quick Reply library

AI matches against your Quick Replies before generating new text. A larger library means more consistent, pre-approved responses and less reliance on generated content.

Use the right mode for the job

Quick grammar fix? Polish. Rough notes into a full response? Transform. No idea what to say? Suggest. Each mode is optimized for its use case.

Set Custom Instructions to reduce edits

If your agents keep editing the same things out of AI responses, add a Custom Instruction. "Never mention competitor products" or "Always include our refund policy link" saves time on every interaction.

Rate limiting is per company

AI requests are throttled to 1 per 5 seconds per company to manage costs. If a request is in progress, wait for it to complete. BYOK users get independent rate limits based on their API plan.

Write Better Replies with AI