Flint AI — Autonomous AI agent for Zendesk that doesn't break the bank
Flint AI is a Zendesk marketplace app that acts as a fully autonomous support agent and an in-ticket co-pilot for your human agents. Built for teams of 1–50 agents who want real AI automation without enterprise-level costs. Install in minutes, train in seconds, go live the same day.
Built for teams that don't want to overpay
Fraction of the cost
From €49/month vs tens of thousands per year for Zendesk Advanced AI. Pay by consumption — not per resolution, not per seat.
Train on the fly
Use /prompt and /sys directly in the playground to update the agent's behavior without leaving Zendesk.
Choose your LLM
Pick GPT-5 Mini (recommended baseline), GPT-5.2, GPT-5.4, Gemini 3.1 Pro, and more. You decide how smart — and how economical — you want it to be.
Live backend integration
Connect Shopify, Magento, custom APIs, or any backend via MCP so the agent can look up orders, subscriptions, and accounts in real time.
Secure and compliant
ISO 9001, ISO 27001, and GDPR compliant. Your knowledge base is private to your instance. Your data is never shared.
Agent + co-pilot in one
Works as an autonomous agent that replies to tickets AND as a sidebar assistant that reads ticket context and helps your agents respond faster.
Everything in one plan — no feature gating
From install to live — step by step
Follow these steps in order. The whole setup takes under 30 minutes for most accounts. Start with internal note mode on so you can test safely before any reply goes public.
Find Flint AI on the Zendesk marketplace
In Zendesk, go to Admin Center → Apps & Integrations → Support Apps. Search for Flint AI and open the product page.
Install and select your domain
Click Install on the marketplace listing. Choose the Zendesk subdomain you want to install it on, then click Next → Install.
Choose a plan and start your free trial
Start with Flint Base — 7-day free trial, all features included, all LLM models available. You won't be charged for 7 days.
Authorize Zendesk access
Inside Zendesk, click Authorize on the yellow banner, then allow access on the Zendesk OAuth screen. This creates the secure connection from your Zendesk to Flint's servers.
Set company name and description
Go to the Agent tab. Add your company name and a short description. This gives the agent context about what it represents and keeps it on topic.
Review the agent instructions
A default CX-optimized prompt is pre-loaded. You can leave it as-is initially or edit it directly. This is the agent's brain — persona, tone, escalation rules, and behavior all live here.
Choose your LLM model
Go to the Config tab. Select your model. GPT-5 Mini is the recommended baseline — most economical for the ticket volume. Use GPT-5.2 or Gemini 3.1 Pro for more complex use cases — they use credits faster.
Set reply mode and answer-as agent
In Config, select which Zendesk agent the bot replies as. Toggle Internal notes only ON to test safely before going live. Turn it OFF when you're ready for public replies.
Upload documents
Go to Data Stores → Upload Documents. Add PDFs, SOPs, product catalogues, internal procedures. Organize by creating folders (e.g. Billing, Technical, Legal).
Index your help center
Click Fetch New Data → Zendesk Help Center. Enter your help center subdomain, click Verify → Next → Start Sync. Works with public articles. Tick "I own this domain" to also pull in private articles.
Index your website (optional)
Use the WordPress / e-commerce source to index your product catalog and pages. Flint indexes images too — the agent can respond with product photos and prices.
Set up data mapping (optional)
Go to Data Mapping. Create rules like: if brand = X, only use content from folder Y. Supports brand, organization, tag, language, and Zendesk form as conditions.
Open MCP Sources
Go to Data Stores → MCP Sources. This is where you connect live backends so the agent can look up real data — orders, subscriptions, accounts, and anything else your API exposes.
Upload your OpenAPI spec
Ask your dev for the OpenAPI spec file — every modern backend has one. Upload it. Flint parses all available API methods automatically.
Select methods and paste your endpoint URL
Choose which API methods to expose (e.g. GET /subscriptions). Paste your endpoint URL. Click Register and Use.
Check tool descriptions
Each connected method comes with an auto-generated description. Review them — the agent uses these to decide when to call which tool. Good descriptions lead to accurate lookups.
Create the trigger
Go to Triggers in the Flint sidebar. Click Create trigger → Yes, I agree. Flint creates the trigger and webhook in your Zendesk automatically. Activate it with the toggle.
Understand the default logic
Conditions: ticket channel is not messaging + status is new or open. Action: notifies Flint via webhook, then solves the ticket (public reply mode) or adds a tag (internal note mode). Optionally add an action at the bottom: ticket status → solved.
Customize for phased rollout (optional)
Edit the trigger to add conditions like tag contains [test-ai] to limit the agent to a subset of tickets during rollout. Keep the trigger inactive if you don't want the agent replying publicly or posting internal notes yet.
Connect the chat widget
Go to Chat Widgets. Click Autoconfig — Flint auto-connects to your Zendesk messaging API. Or manually: create an API key in Zendesk Admin Center and paste the App ID, Key ID, and Secret Key.
Customize bot name and greeting
Give your bot a name and set an opening message. Choose whether to display an "AI" label next to responses. Then select which widget to link.
Embed on your website
Go to Zendesk Admin Center → Channels → Messaging → your widget → Edit. Copy the embed code and paste it into your website's page HTML (e.g. Squarespace page settings → Advanced → Header Code Injection).
Use the Playground
Go to Playground. This is your sandbox — chat with the agent as if you were a customer. Select your test model from Quick Settings. What you test here is separate from what the live agent uses to reply.
Interrogate behavior with /sys
Type /sys followed by your question (e.g. "why didn't you include product links?"). A smart LLM analyzes the agent's reasoning against its current prompt and tells you exactly why it answered the way it did.
Update the prompt with /prompt
Type /prompt followed by your instruction (e.g. "next time include a product URL for any recommendation"). Flint generates a diff — green = added, red = removed. Review and save as a new version.
Check RAG results with Debug
After any Playground reply, click Debug. You'll see which knowledge base articles were retrieved, their relevance score (0–1), and how long each step took. Low scores = the knowledge base needs better content for that topic.
Manage prompt versions
All saved prompt versions appear in the Prompts panel on the right. The active version has a green dot. You can switch back to any previous version instantly.
Open any ticket in Zendesk
Flint AI appears as a sidebar panel on the right of every ticket. It reads the full ticket thread automatically — no setup needed.
Use quick actions
Copilot — reads the ticket, checks your SOPs and knowledge base, and tells the agent what to do. Summarize — condenses long threads. Suggest answer — drafts a ready-to-send reply based on ticket context.
Chat freely in the sidebar
Ask anything in plain text: "translate this to German", "what's the refund policy for this product", "should I escalate this?". The co-pilot uses the same knowledge base as the autonomous agent.
If you feel stuck,