CS Agent — AI-Powered Customer Service

How to use CS Agent

Setup guide & provider comparison

Quick start

Choose your AI Provider from the left sidebar and paste your API key. It saves automatically.

Paste the full customer email thread — not just the latest message. The more history, the better the reply.

Add context — order numbers, account info, prior complaints, product details. Attach screenshots or photos if relevant.

If you've set up a Knowledge Base platform (links, docs, or notes), that info is not used automatically. You must select that platform's chip in the platform bar above the message box (Draft tab) or conversation box (QA tab). If nothing is selected, the AI won't see it and may give a generic or "wrong" answer — even though the info is saved.

Hit Generate response (or press Ctrl+Enter) and copy the drafted reply.

Understanding the tabs

Draft Response

Where you write replies. Paste the customer's message and any context, then generate a ready-to-send draft. If you've added a platform in the Knowledge Base, use the platform bar just above the message box to select it — that's what tells the AI to actually use that platform's saved policies and info.

QA Review

Where you check a conversation after the fact — your own reply, or one written by another agent. Paste the full back-and-forth thread and the AI scores it for accuracy, tone, completeness, and risk, then flags specific issues. Just like the Draft tab, select the relevant platform chip here so the AI checks the reply against that platform's actual policies instead of general best practices.

Knowledge Base

Where you store reference info per platform (e.g. Shopify, Amazon, TikTok Shop) — reference links, uploaded documents, and text notes like policies or FAQs. Adding info here only saves it — it doesn't automatically get used. You still need to go to the Draft or QA tab and select that platform's chip in the platform bar for the AI to actually read it.

AI Providers — which one to choose?

🟣 Anthropic — Claude

Paid — free trial credits

The most capable option for customer service work. Excellent at reading long conversation histories, analyzing images (screenshots, product photos), and writing nuanced, context-aware replies. Recommended if you want the best quality responses.

Get API key — console.anthropic.com View pricing

💡 New accounts get free credits to start. After that, pay-as-you-go. Claude Sonnet 4 is the sweet spot of cost vs quality.

🔵 Google — Gemini

Free tier available

Google's AI with a generous free tier. Gemini 2.0 Flash is fast and capable — great for most customer service tasks. Free usage has rate limits (requests per minute/day) but is more than enough for testing and low-to-medium volume use.

Get API key — aistudio.google.com View pricing

💡 Gemini 2.0 Flash is free up to 1,500 requests/day — ideal for getting started at zero cost.

⚡ Groq

Free tier available

Groq runs open-source models (Llama, Mixtral) on custom hardware — the result is extremely fast responses, often 5–10× faster than other providers. Great for quick drafts. Free tier has daily token limits. Uses open-source models which are slightly less polished than Claude or Gemini for nuanced writing.

Get API key — console.groq.com View pricing

💡 Best if speed is the priority. Llama 3.3 70B is the recommended model — open-source but surprisingly capable for CS work.

🔀 OpenRouter

Free models available

A marketplace that gives you access to many AI models through a single API key — including free community-hosted models. This app currently uses only free models for OpenRouter, making it perfect for testing at zero cost. Quality varies per model; GPT-4o Mini is a reliable starting point.

Get API key — openrouter.ai/keys Browse free models

💡 Sign up is free, no credit card needed for free models. Good for exploring different models before committing to a paid provider.

Quick comparison

Provider	Cost	Speed	Quality	Images	Best for
🟣 Anthropic	Paid	Fast	★★★★★	✓ Full	Production use
🔵 Gemini	Free tier	Fast	★★★★☆	✓ Limited	Free daily use
⚡ Groq	Free tier	⚡ Fastest	★★★☆☆	✗ No	Speed priority
🔀 OpenRouter	Free models	Varies	★★★☆☆	✗ No	Testing & exploring

Tips for better responses

📋 Paste the full email thread — not just the latest message. Include every previous reply so the AI sees all prior commitments and won't contradict them.

🖼 Upload product photos or screenshots — if the customer sent an image of a "defect", paste it here. The AI will analyze it and may find it's normal behavior, saving you a wrong escalation.

📄 Paste your policies in the constraints field — refund windows, escalation rules, what's covered under warranty. The AI will work within them instead of making promises you can't keep.

🏬 Select the right platform chip — if you've built up a Knowledge Base for a platform, click its chip in the platform bar on the Draft or QA tab before generating. Without this step, that saved info is ignored, even though it's stored.

🔐 Your API key stays on your device — it is saved in your browser's local storage and is never sent to any server other than the AI provider you choose.

Settings

AI Provider

API Key

Get your key from console.anthropic.com

Session only — forget this key when the tab closes (safer on shared/public computers)

Set a spending cap. A browser-stored key can't be made 100% unstealable, so cap the damage at the provider instead: console.anthropic.com/settings/limits — set a low monthly budget so a leaked key can only cost a few dollars, not your whole balance.

Model

Response tone

Response language

Response length

Company / brand name

Rep name (optional)

Department

Your API key is used directly in the browser and never sent to any server other than the provider you select.

Customer message & context

Customer email / message *

Conversation history & context

What was already said or previously known. Need a product page, manual, or FAQ link? Add it once in the Knowledge Base tab so it's reused on every future ticket for this platform.

Order / account status (optional)

Current facts from your system, not conversation — treated as ground truth, and checked against the reply in the grounding checklist.

Policies / constraints

For standing, reusable policies (refund window, warranty terms), add them once in the Knowledge Base tab instead — this field is for one-off exceptions specific to this ticket.

Attachments (images for context)

Drop images / screenshots here or click to browse

Urgency level

0 chars

Drafted response

Your AI-drafted response will appear here.
Fill in the customer message and hit Generate response.

Use the 💬 bubble (bottom-right) to refine with AI.

QA Settings

Provider & model

Uses the same provider and API key as the Draft tab.

Evaluation strictness

Focus areas

Accuracy & policy compliance Completeness Tone & empathy Resolution & next steps Agent consistency Legal / escalation risk

Company policies (optional)

Without this, AI judges against general best practices.

Conversation to review

Full conversation thread *

The more complete the thread, the more accurate the review.

Additional context (optional)

0 chars

Review results

Paste a conversation and press Run QA Review.