AI settings

Customize how StarkoAi behaves and responds, and connect tools and custom actions to achieve more.

Articles

AI Settings: Customize and Control Your Assistant

The AI Settings tab in Starko.one is where you take full control over how your AI assistant behaves, responds, and integrates with your workflows. This is where your assistant stops being generic—and starts acting as a true extension of your brand. From defining personality and tone, to integrating backend systems and automating custom actions, the AI Settings tab empowers you to fine-tune every aspect of your assistant’s performance to better reflect your company’s voice, goals, and customer expectations. Personalizing the Assistant The assistant can be hand-tailored to match your brand identity and communication style. You can modify: Tone of Voice – Choose whether your assistant is formal, friendly, technical, casual, or anything in between. Response Length – Set whether answers are concise, detailed, or adaptive based on context. Personality – Customize the assistant’s overall persona (e.g., helpful and professional, witty and playful, or warm and empathetic). This personalization ensures a consistent customer experience aligned with how your brand communicates. Company Profile & Context To help your assistant provide more accurate, relevant answers, you can input your: Company name and mission Product or service descriptions This information helps the assistant align more closely with your operational context and provide clients with accurate, business-specific answers. System Prompt Customization For advanced configuration, the AI Settings tab lets you edit the system prompt — the underlying instruction set that governs how the assistant reasons and responds. You can: Define your assistant’s role and limits Outline preferred behavior for different use cases Include example conversations or domain-specific instructions This is especially useful for highly regulated industries, complex products, or teams that want granular control over AI behavior. Behavior Control & Escalation Rules You can specify what the assistant should do in different types of interactions: Escalate to a human agent when confidence is low, sentiment is negative, or the topic exceeds the assistant’s scope. Open a support ticket automatically based on conversation triggers or requests. Continue autonomously if the interaction falls within predefined comfort zones. This flexibility ensures the assistant knows when to handle issues on its own—and when to get out of the way. Custom Actions & Workflow Integration Starko allows you to define Custom Actions that your assistant can execute during a conversation. These can be linked to: Internal tools (e.g., CRMs, inventory systems, scheduling platforms) External APIs (via webhooks) Business logic (e.g., cancel a subscription, look up a shipping status, trigger a Slack alert) Examples of custom actions include: “Check order status in Shopify” “Create a task in Asana” “Send a password reset email” This makes your assistant more than a support bot—it becomes an active part of your operations. Customer Feedback Monitoring Within the AI Settings tab, you can monitor customer feedback on the assistant’s performance in real time. Key features include: Rating collection (thumbs up/down or custom metrics) Feedback tagging (e.g., “unhelpful,” “too long,” “too vague”) Sentiment analysis to detect frustration, confusion, or satisfaction This feedback loop enables continuous learning and refinement, so your assistant becomes smarter, more accurate, and more aligned with user needs over time. Conversation Review & Performance Analytics Gain deeper insights into how your AI is performing with tools to: Browse recent conversations Identify escalation triggers Flag missed opportunities or failed resolutions Track usage trends and resolution rates By analyzing these interactions, you can spot gaps in knowledge, refine tone settings, and proactively improve your AI’s behavior. Why AI Settings Matter The AI assistant in Starko.one is only as effective as it is aligned with your company. The AI Settings tab ensures your assistant: Speaks with your voice Acts according to your rules Grows with your business Integrates into your workflows Learns from real-world feedback It transforms your assistant from a static responder to a dynamic, evolving part of your support and operations team. Getting Started with AI Settings Navigate to the AI Settings tab in your Starko dashboard. Configure the assistant’s tone, response style, and personality. Fill out your company profile and contextual data. Customize the system prompt for deeper control. Define escalation rules and set up any required custom actions. Monitor customer feedback and review performance regularly to refine and adapt. image how to create custom actions:

Procedures: build step-by-step playbooks for your AI agent

Procedures let you teach your AI agent to handle specific requests with a\ repeatable, multi-step flow — the same way you'd train a new teammate. When a\ customer's message matches what a procedure is for, the agent follows your steps\ in order: asking questions, looking things up, using your tools, getting a\ teammate's approval when needed, and resolving the request. This guide covers what Procedures are, how to build one, and how they behave in a\ live conversation. When to use a Procedure Your AI agent already answers questions on its own from your knowledge base. Use\ a Procedure when a request needs a specific sequence of actions rather\ than a single answer — for example: Processing a refund (verify the order → check eligibility → refund or explain) Quotes or plan recommendations (collect details → calculate → recommend) Anything that needs a teammate's approval before acting Anything that collects information across a few messages If a request can be answered in one reply from your help articles, you don't need\ a procedure — let the agent handle it normally. Create a Procedure Open Procedures in the sidebar and click New procedure. Name it clearly (e.g. "Refund request"). Describe when it should run. In the trigger field, write the customer\ intent that should start it, in plain language — like "The customer wants a\ refund or to return an order." The agent matches on meaning, so you don't\ need exact keywords. Add your steps (see the building blocks below). Save, then turn it on to make it live. While it's off it stays a draft\ and won't run. 💡 Start simple. A few clear steps work better than a long, deeply branched flow. The building blocks Add steps from the + Add step menu. Each block does one thing: Block What it does --- --- Step A plain instruction in your words ("Greet the customer and confirm the issue"). The agent turns it into a natural message or action. If / Else Branches based on a condition. Steps under If run when it's true; Otherwise runs when it's not. Wait for approval Pauses until a teammate approves or rejects, then runs the If approved / If rejected branch. Use a tool Calls one of your connected tools — a custom action, an MCP tool, or a built-in (create ticket, search knowledge base, etc.). Ask the customer Asks a question and waits for the reply before continuing. Look up content Searches your knowledge base for relevant articles. Hand off to human Connects the conversation to a teammate and ends the procedure. End Finishes the procedure with an optional closing message. Using your tools inside a step While writing a step, type to mention a tool — your custom actions, MCP\ tools, and built-ins appear in a list. Mentioning a tool lets the agent use it\ for that step only, so it can't reach for anything it shouldn't. How a Procedure behaves in a conversation It starts automatically. When a customer's message matches a procedure's\ trigger, the agent begins following it — no command needed. One at a time. A conversation runs one procedure at a time. If the customer\ clearly switches to a different request, the agent switches too. It asks and waits. On an Ask the customer step, the agent sends the\ question and waits — even if the customer replies a while later, it resumes\ right where it left off. It can pause for your approval. On a Wait for approval step, an\ Approval needed prompt appears on the conversation with Approve /\ Reject buttons. The procedure continues down the matching branch once you\ decide; the customer is told you're checking in the meantime. It handles side questions. If the customer asks something unrelated mid-flow\ ("by the way, what are your hours?"), the agent answers it from your knowledge\ base, then steers back to the procedure — without losing its place. It speaks naturally. Replies are written in the customer's language and your\ brand's tone. Your step text is guidance for the agent, not a script it reads\ aloud. It ends cleanly. When the flow reaches End (or hands off), the\ conversation wraps up with a clear message. Example: a refund procedure Ask the customer — "What's the order number you'd like refunded?" Use a tool — look up the order. If / Else — If the order is within 30 days: Wait for approval — "Approve refund for this order?" If approved: Use a tool to process the refund → Step: confirm to\ the customer. If rejected: Step — explain it can't be refunded and offer options. Otherwise: Step — explain the 30-day policy. End — "Thanks for reaching out — anything else I can help with?" Tips & best practices Write triggers by intent, not keywords — describe what the customer wants. Keep steps focused — one action or message per step is clearer and more reliable. Use approvals for anything sensitive — refunds, account changes, etc. Only mention the tools a step needs — it keeps each step precise. Test it in your widget before turning it live for everyone. Quick start: Create your first Procedure Procedures are step-by-step playbooks your AI agent follows to resolve specific customer requests — booking a demo, checking an order, processing a refund. When a conversation matches a procedure's trigger, the agent works through your steps top to bottom, pausing to ask the customer, calling tools, branching on conditions, or handing off to a teammate exactly where you tell it to. This guide walks you through building one from scratch in about five minutes, using a "Book a demo" example. The Procedures list — click New procedure (top right) to start. Create the procedure In the left sidebar, go to Procedures. Click New procedure (top right). You'll land straight in the procedure editor with an "Untitled procedure". Give it a clear name at the top — e.g. Book a demo — so your team can find it later. The procedure editor — name at the top, the Draft badge, and the When to use this procedure and Steps sections. At the top you'll also see a Draft badge. A procedure only runs once it's set Live — we'll do that at the end, after it's built and tested. Define when it should run The When to use this procedure box is how the agent decides whether a conversation matches. Describe the customer intent in plain language — and, just as importantly, when it should not fire. For our example: Trigger when the customer wants to book or schedule a product demo. > Only trigger if the customer: > - asks to book, schedule, or set up a demo or call > Do not trigger if: > - they only ask about pricing or features they already have a demo booked and want to reschedule Tip: Be specific about the "do not" cases. The clearest line you can draw is the one between this procedure and the next one, so the agent doesn't fire the wrong playbook. (The filled-in trigger box is shown in the editor screenshot above — it grows to fit as you type.) Add the steps This is the heart of the procedure. Click + Add step to open the step menu — each step type does a specific job: Step type What it does --- --- Step A plain instruction the agent follows. If / Else Branch on a condition (e.g. if the slot is available…). Wait for approval Pause for a teammate to approve before continuing. Use a tool Call one of your connected tools (calendar, CRM, internal API). Ask the customer Ask a question and pause until they reply. Look up content Search your Knowledge Hub for an answer. Hand off to human Route the conversation to a teammate (ends the procedure). End Close the procedure with an optional final message. The + Add step menu — every step type in one place. Steps run top to bottom. You can reorder or delete any step with the controls that appear on hover. Build the "Book a demo" flow Step 1 — Ask the customer. Add an Ask the customer step: What day and time works best for your demo? This is the key one to get right: an Ask the customer step pauses the procedure until the customer replies. (A plain Step that says "ask them…" does not pause — the agent keeps going. When it matters that the agent waits, always use an Ask the customer step.) An Ask the customer step — note the "pauses until they reply" label. Step 2 — Use a tool. Add a Use a tool step, select your calendar/booking tool, and describe what to do: Check availability for the requested time and book the demo. You can also reference a tool inline inside any step's text with Insert tool (the @ button) — handy when a single instruction touches more than one tool. A Use a tool step — pick a connected tool, then tell the agent what to do with it. Step 3 — If / Else. Add an If / Else step to handle the two outcomes: Condition: the requested time is available IF: Confirm the booking and tell the customer they'll receive a calendar invite. OTHERWISE: Offer the three nearest available times and ask them to pick one. (The completed If / Else step, with its IF and OTHERWISE branches, is shown in the full procedure below.) Step 4 — End. Add an End step with a friendly closing message: Great — your demo is booked! You'll get a calendar invite shortly. Anything else I can help with? The full Book a demo procedure — Ask → Use a tool → If / Else → End. Writing effective steps A few habits that make procedures reliable: Write for a smart new teammate. Plain, direct instructions ("Confirm the booking and share the invite") beat vague ones ("handle the booking"). One action per step. Smaller steps are easier for the agent to follow and easier for you to debug. Use the right pause. Ask the customer and Wait for approval are the only steps that pause — reach for them whenever the flow must stop and wait. Branch on real signals. In an If / Else condition, point at something concrete the agent can know (a tool result, the channel, the order age) rather than a vague judgment call. If you write a question into a plain Step, STARKO flags it inline — a reminder that plain steps don't wait for a reply: Pro tip: Need a human in the loop for anything sensitive (issuing a refund, sending a contract)? Drop a Wait for approval step in front of it — the procedure pauses until a teammate approves, then continues down the IF APPROVED branch. Set it live When you're happy with the flow: Click Save. Click Set live. The badge flips from Draft to Live, and the agent will start using the procedure on matching conversations. To pause it at any time, open it and click Set to draft — your steps are kept, it just stops firing. Save your work, then click Set live. The badge flips from Draft to Live. FAQ When does a procedure actually run?Only when it's Live and a conversation matches the When to use this procedure description. If two procedures could match, tighten the "do not trigger" lines so the agent has a clear choice. Why didn't the agent wait for the customer's answer?A plain Step never pauses. If the agent should stop and wait for a reply, use an Ask the customer step; for teammate sign-off, use Wait for approval. Can one step use more than one tool?Yes — add a Use a tool step for the primary tool, and reference any others inline with Insert tool (@) inside the step text. What happens when a procedure ends?A Hand off to human step routes the conversation to your team and ends the procedure. An End step closes it with an optional final message. If neither is reached, the agent simply finishes the last step and continues the conversation normally. Next: connect your tools so procedures can take real actions, and explore branching with If / Else and Wait for approval for more advanced flows. FAQ Do Procedures replace the AI's normal answers?No. The agent still answers everyday questions from your knowledge base.\ Procedures only start for the specific requests you've defined. What if the customer goes off-script?The agent answers their side question, then guides the conversation back. If\ they've clearly moved to a different request, it switches procedures. Which channels do Procedures work on?Web chat, email, WhatsApp, and Instagram. What happens if no one approves a "Wait for approval" step?The procedure stays paused and the customer is kept informed. It resumes the\ moment a teammate approves or rejects.