How to Draft Front Replies with AI (From Your Knowledge Base)

What "AI Compose" actually means in Front

When people say they want to draft Front replies with AI, they almost always mean Copilot — Front's in-inbox writing assistant. Per Front's guide to jumpstarting replies with Copilot, you invoke it two ways: hit the spacebar when the composer is empty, or click the Compose with Copilot icon in the reply box. From there you can draft a fresh reply from a short prompt, or take text you've already written and refine it.

The refinement side is genuinely useful. Copilot offers preset options — make it more concise, add a friendlier tone, turn a wall of text into numbered steps, summarise a long thread — alongside a free-text field for custom prompts. You're not stuck with the first draft; you iterate until it reads right, then you review and send it yourself. Copilot generates drafts for human approval — it does not auto-send in the composer.

Crucially, the quality of that draft depends entirely on what Copilot can see. Out of the box it leans on past, similar conversations in your Front history. But to have it answer from your documented policies and product facts, you have to connect your knowledge base — which is the difference between a draft that guesses and a draft that's grounded.

Step-by-step: draft a grounded reply

Here's the practical sequence a team follows to get from an incoming message to a reviewed, on-brand reply.

1. Connect your knowledge sources first. In Front AI settings, add the sources Copilot should draw from. Per Front's guide to maximizing AI with knowledge sources, you can connect Front Knowledge Bases (help articles, updated to AI within minutes), Public Websites (synced by URL, up to 3,000 pages crawled), third-party/private sources, file uploads (PDFs up to 25 MB each, 500 MB per workspace), short-form Notes, and your historical conversations. Until the knowledge base is connected in settings, Copilot can't quote your articles — it only has similar past conversations to lean on.
2. Open the conversation and trigger Copilot. Spacebar on an empty composer, or the Compose with Copilot icon.
3. Give it intent, not boilerplate. A short prompt like "reassure this customer their delayed order is on the way and give the new ETA" beats pasting a template — Copilot pulls the relevant facts from your connected sources.
4. Refine for tone and shape. Use the preset options (more concise, friendlier, bulleted steps) or a custom prompt to match your brand voice before anything leaves the inbox.
5. Drop in a message template where it fits. Front's saved message templates handle the fixed scaffolding — greeting, signature, legal footer — while Copilot fills the situation-specific middle. The two are complementary, not competing.
6. Review, then send. A human reads the draft, checks the facts, and sends from the shared inbox. Nothing goes out unread.

One operational note worth planning around: Copilot is rate-limited to 200 requests per user per day, where each compose or transformation counts as a request. For most agents that's plenty; for a high-volume queue, it's a ceiling to be aware of.

(The screenshot shows a Macha demo-org agent — the Customer Reply Assistant — demonstrating the grounded-draft capability. Front isn't a connected connector in this demo org, so this isn't a live Front conversation; it illustrates the same KB-grounded draft-a-human-sends pattern that Macha runs on top of Front.)

Keeping drafts on-brand: templates and tone

A shared inbox lives or dies on consistency. Ten agents answering the same billing question shouldn't produce ten different voices, and this is where AI compose needs guardrails.

Message templates carry the structure; the knowledge base carries the truth. Front's templates are the reliable scaffolding — the way you open, the way you sign off, the disclaimers legal insists on. Let those stay fixed. Point Copilot at the situation-specific part: the actual answer to this customer's this question. When you refine, use the tone presets deliberately so the middle matches the scaffolding around it.

Ground everything in a source of truth. Front itself frames this well: AI is good at language, but "it needs a source of truth to provide accurate, helpful, and brand-consistent answers." A draft grounded in your KB article about refund windows will use your refund window — not a plausible-sounding number the model invented. That's the whole game with AI compose: grounding is what separates a helpful draft from a confident liability.

Where native Front AI stops — honestly

Front's AI is not a token feature, and it's worth crediting properly. Copilot genuinely speeds up drafting from your knowledge base and history. Autopilot goes further — an omnichannel AI agent that can auto-reply to conversations based on Topics and run multi-step customer work. And Smart QA scores conversations for quality at scale. For a lot of teams, that's a real toolkit.

But there are edges worth naming plainly, because the honest version helps you plan:

• Compose drafts; it doesn't resolve. Copilot writes the words. It doesn't, on its own, look up whether this customer's order actually shipped, check their invoice, or take an action in another system. The reasoning-and-doing part — the part that turns a good draft into a genuine resolution — is where in-inbox compose reaches its natural limit.
• The good stuff is add-on billing. Per front.com/pricing, Autopilot is a paid add-on "starting at $0.05 / conversation," Copilot and Smart QA are each around $20/seat/mo, and while these are bundled into Enterprise, Starter and Professional teams pay for them on top of per-seat plans. Front's pricing tiers show where each unlocks. That's not a criticism — it's a line item to budget for.
• AI Answers is retired. Front's older AI Answers chatbot is now legacy: the help center states it is "no longer available for purchase," and Front directs new buyers to Autopilot Resolve instead. If you read older comparisons still citing AI Answers, they're describing a product you can no longer buy.

Here's a way to think about the cost model, since it's the question that trips teams up most:

The distinction in the right column matters: Macha's credits are consumed per AI action — never "per resolution." Automation, reasoning, and a tool call are different amounts of work, and it's more honest to meter the work than to bill a fuzzy outcome.

Where a grounded AI agent layer picks up

This is the seam. Native compose is excellent at writing; it's not built to reason over live data and act. That's the job of a dedicated AI agent — and the wider category of AI agents for customer service exists precisely for the work an in-inbox writer can't finish.

Macha is one such layer, and the positioning matters: it runs on top of the Front you already use through the Macha–Front connector — it does not replace Front, your shared inboxes, your templates, or Copilot. You keep drafting in the inbox exactly as you do now. Where Macha adds edge is grounding plus reasoning plus action: its agent reads the intent of a conversation, pulls a real order status or account detail through a custom tool that turns your REST API into something the agent can call, and drafts a reply your teammate reviews and sends from the shared inbox — or, once you trust it, sends directly.

That's the difference between "here's a nicely worded draft" and "here's the actual answer to this specific customer, with their real ETA in it." If you want the mechanics, connecting Front to Macha to route conversations to AI and the guide to building an AI agent for Front both walk through it, and Front AI explained covers the native feature set in full.

The clean division of labour: let Copilot and templates handle fast in-inbox drafting, and layer a grounded agent on top for the reasoning-and-doing that a writing assistant structurally can't reach.

FAQ

How do I draft a reply with AI in Front? Open a conversation, then press the spacebar in an empty composer or click the Compose with Copilot icon. Give Copilot a short prompt describing what you want to say, refine the draft using the preset tone options or a custom prompt, then review and send it yourself. Copilot drafts for human approval — it doesn't auto-send from the composer.

Does Front Copilot use my knowledge base? Only once you connect it. Copilot draws on past similar conversations by default, but to draft from your documented articles you must add your knowledge base in Front AI settings. Front supports Front Knowledge Bases, public websites (up to 3,000 pages), third-party sources, file uploads, notes, and historical conversations.

What's the difference between Copilot and Autopilot? Copilot is an in-inbox writing assistant that drafts replies for a human to review and send. Autopilot is an omnichannel AI agent that can auto-reply to conversations and run multi-step work. Both are paid add-ons on Starter and Professional (bundled into Enterprise), per front.com/pricing.

Is Front AI Answers still available? No. AI Answers is a legacy feature that Front's help center says is "no longer available for purchase." Front now directs new buyers to Autopilot Resolve for automated AI replies.

Can I add a grounded AI agent to Front without replacing it? Yes. An AI agent layer like Macha connects to Front as a live connector and runs on top of your existing inboxes, templates, and Copilot — it doesn't replace them. Your team keeps drafting in the inbox; the agent adds intent-reading, live data lookups via custom tools, and grounded drafts a teammate reviews and sends.

Ready to turn a good draft into the actual answer? Start a free trial of Macha and connect it to your Front in minutes.