> ## Documentation Index > Fetch the complete documentation index at: https://docs.runspark.ai/llms.txt > Use this file to discover all available pages before exploring further. # Chatting with Your Agent > How to talk to your agent, what to ask, and how to teach it through conversation Your agent lives in the chat panel on the right side of your Spark workspace. Click on your agent in the sidebar, and you'll see a text box where you can type messages — just like texting a coworker. **Here's the most important thing to know: just talk naturally.** You don't need special commands, keywords, or formatting. Type like you're messaging a colleague, and your agent will understand. *** ## Your First Conversation If you're staring at the chat panel thinking "what do I even say?" — start here. Just type one of these: > **You:** Hi! What can you do? Your agent will introduce itself and explain its capabilities based on how it's been set up. > **You:** I'm new to this. Can you walk me through how to use you? Your agent will give you a personalized tour based on what's been configured in your workspace. > **You:** What do you know about my business so far? This is a great way to see what knowledge files have been uploaded and what your agent already knows. If the answer is "not much yet," that's your cue to start teaching it. **You literally cannot break anything by typing the wrong thing.** There's no "wrong" message. If your agent misunderstands you, just clarify. If it does something you didn't want, just say "undo that" or "that's not what I meant." Think of it like talking to a new employee — they might not get it right the first time, but they'll learn fast. *** ## How to Talk to Your Agent Effectively Your agent is smart, but it's not a mind reader. The more context you give, the better the results. Here's the difference: ### Be Specific **You:** Look up the contact Sarah Johnson, check when we last spoke with her, and draft a follow-up email offering our premium package with the Q2 pricing. **Agent:** Found Sarah Johnson — your last interaction was a phone call on March 12th where she expressed interest in upgrading. Here's a draft follow-up email: *"Hi Sarah, great speaking with you last week about upgrading to our premium package. As we discussed, the Q2 pricing is \$3,200/month for your team size, which includes priority support and advanced analytics. I'd love to set up a quick call to finalize the details. What does your Thursday look like? Best, \[Your name]"* Want me to adjust anything before sending? **You:** Email Sarah. **Agent:** I have a few contacts named Sarah. Which one do you mean? Also, what would you like the email to say? ### Give Context Your agent knows what's in its knowledge files and memory, but it doesn't know what just happened in your day unless you tell it: > **You:** I just got off a call with the Martinez family. They want to book a birthday party for 30 kids on April 20th. They asked about the deluxe package but are concerned about the cost. Can you generate a quote and draft a response that highlights the value? With that context, your agent can pull up the right pricing, create a tailored quote, and draft a response that addresses their specific concern about cost. Without it, your agent would have to ask you five follow-up questions. ### Tell It What You Want the Output to Look Like > **You:** Give me a summary of this week's new leads. Keep it to bullet points — name, what they're interested in, and how hot the lead is. No more than one line per lead. Compare that to: > **You:** Show me the leads. Both work, but the first one gets you exactly what you need in the format you want, while the second might give you a long table when you just wanted a quick list. *** ## Teaching Your Agent Through Conversation One of the most powerful things about your agent is that you can teach it just by talking to it. No settings pages to configure, no forms to fill out — just explain things naturally. ### Setting Rules > **You:** New rule: whenever a lead asks about custom pricing, always ask them for their team size and expected start date before giving a quote. > > **Agent:** Rule added! From now on, when leads ask about custom pricing, I'll always ask for their team size and expected start date before generating a quote. ### Correcting Mistakes When your agent gets something wrong, tell it. This is how it learns: > **You:** That quote is wrong. The seasonal discount only applies to bookings in June through August, not all summer long. Fix that and remember it. > > **Agent:** You're right — I've corrected the quote to apply the seasonal discount only for June through August bookings. I've updated my rules so I won't make that mistake again. Here's the corrected quote: ... ### Updating Preferences > **You:** I don't like how you format reports. Use this format going forward: start with a one-sentence summary, then bullet points for key metrics, then a "needs attention" section at the bottom. > > **Agent:** Got it! I've updated my report formatting. From now on, I'll use: 1) One-sentence summary up top, 2) Bullet points for key metrics, 3) "Needs attention" section at the end. Want me to redo the last report in this new format? ### Defining Processes > **You:** Let me walk you through how we handle a new client signup. Step 1: create their contact record with all their info. Step 2: send them our welcome email template. Step 3: create a task for me to do a kickoff call within 48 hours. Step 4: add them to the "Onboarding" stage in the pipeline. Got it? > > **Agent:** Got it! Here's the new client signup process I've saved: > > 1. Create contact record with all provided information > 2. Send the welcome email template > 3. Create a task for you to do a kickoff call within 48 hours > 4. Move them to the "Onboarding" pipeline stage > > I'll follow these steps whenever we have a new client signup. Want me to run through this process for anyone right now? Learn more about defining processes in our [dedicated guide](/guides/defining-processes). *** ## Common Things to Ask Your Agent Not sure what to type? Here are real examples organized by category. You can copy any of these and use them right now: ### 📋 CRM & Contacts > "Create a new contact: David Kim, [david@techfirm.com](mailto:david@techfirm.com), 555-0198. He's a warm lead interested in the enterprise plan." > "Show me all leads that haven't been contacted in the past 5 days." > "Update the Thompson deal — they've agreed to the proposal. Move it to 'Closed Won' and set the value to \$24,000." > "Who are my hottest leads right now? Show me the top 5 by deal value." > "Add a note to Maria Garcia's contact: she mentioned she's going on vacation until April 10th. Don't reach out until then." > "How many new contacts did we add this month?" ### 📧 Email & Communication > "Draft a follow-up email to everyone in the 'Proposal Sent' stage who we haven't heard from in a week. Keep it friendly, not pushy." > "Check my email — is there anything urgent I need to respond to?" > "Write a cold outreach email for SaaS companies. Tone: professional but warm. Mention our free trial." > "Reply to that email from James and let him know we can schedule the demo for next Tuesday at 3pm." ### 📊 Reports & Analysis > "Give me a weekly pipeline summary — total deals at each stage, total value, and which deals moved this week." > "How did we do last month? I want total new leads, deals closed, revenue, and average deal size." > "Create a report comparing this quarter's performance to last quarter." > "Which lead sources are bringing in the most high-value deals?" ### ⏰ Tasks & Reminders > "Remind me to call the Henderson team on Thursday at 2pm." > "What tasks are due today?" > "Create a task for next Monday: review and update the pricing guide for Q2." > "I need to follow up with every lead who requested a demo. Create tasks for each one, due this Friday." ### 📅 Scheduling & Automation > "Set up a daily task: every morning at 8am, check for new leads and send me a summary." > "Every Friday at 5pm, generate a weekly pipeline report and save it to my Work tab." > "On the 1st of every month, create a monthly performance summary." > "Cancel the daily email check — I don't need that anymore." ### 📄 Documents & Proposals > "Write a proposal for the Westfield account. They want our pro plan for 50 users starting June 1st." > "Create a one-pager about our premium support tier that I can send to prospects." > "Draft a quarterly business review document for the Ramirez account. Pull data from their deal history." > "Rewrite paragraph 3 of that proposal — make it more concise and focus on ROI." ### 🧠 Knowledge & Rules > "What do you know about our cancellation policy?" > "What's our pricing for teams over 100 people?" > "New rule: always include our money-back guarantee when discussing pricing with nervous leads." > "What rules do you currently have for handling new leads?" ### 🏗️ Dashboard & Workspace > "Build me a dashboard showing: new leads this week, total pipeline value, deals closing this month, and our conversion rate." > "How many deals are in the negotiation stage right now?" > "Show me a breakdown of deals by source — referral vs. website vs. cold outreach." ### 🤔 About Your Agent > "What can you do?" > "What do you know about my business?" > "What rules are you currently following?" > "Show me your schedule — what recurring tasks do you have set up?" > "What did we talk about yesterday?" *** ## Correcting Your Agent (This Is Important!) Your agent will make mistakes — especially in the first few days. **This is completely normal** and actually part of the process. Here's how to handle it: ### When the Answer Is Wrong > **You:** That's not right. Our standard package starts at $500, not $400. Where did you get that number? > > **Agent:** You're right — I apologize. I was using an outdated figure. I've corrected my notes: the standard package starts at \$500. I'll use this going forward. The number I had was from an older version of the pricing guide — you may want to check if the knowledge file needs updating. Notice what happened: the agent corrected itself, told you where the mistake came from, and suggested checking the source. If the knowledge file is outdated, update it and your agent will use the new numbers from then on. ### When the Process Is Wrong > **You:** No — we don't send the quote directly. We always schedule a call first to discuss their needs, THEN send a customized quote after. Update your process. > > **Agent:** Updated! New process for quotes: 1) Schedule a call to discuss needs first, 2) After the call, create a customized quote, 3) Send the quote. I won't send quotes directly to leads anymore. ### When the Tone Is Off > **You:** That email was way too formal. We're a small, friendly company — write it like you're talking to a friend, not a corporate client. Keep it professional but casual. > > **Agent:** Got it — adjusting my email style. Here's the rewritten version: > > *"Hey James! Thanks for reaching out — sounds like you're looking for something right in our wheelhouse. I'd love to hop on a quick call to see what would work best for your team. Got 15 minutes this week? — Alex"* > > Better? I'll use this tone for emails going forward. ### When You Want to See What It Learned > **You:** What corrections have you logged from our conversations? > > **Agent:** Here are the key things I've learned from your feedback: > > * Standard package starts at $500 (not $400) > * Seasonal discount only applies June-August > * Always schedule a call before sending quotes > * Use casual-professional tone in emails > * Round up to the next package size when calculating quotes > > Want me to adjust any of these? **Every correction makes your agent permanently smarter.** When you tell it "that's not how we do things," it doesn't just fix the immediate mistake — it writes down the correction and applies it to every future interaction. The 10 minutes you spend correcting your agent in week one saves you hours of corrections in month two. *** ## Conversation History Your chat history with your agent is **persistent** — it's saved and available even after you close your browser. You can scroll back through previous messages anytime. A few things to know: * **Each channel is separate.** Your conversation in the Spark portal is a different thread than your conversation on Telegram or Slack. Think of them like separate chat rooms with the same person. * **Memory is shared across channels.** Even though the conversations are separate, your agent's knowledge and memory carry across everywhere. If you told it something via Telegram, it remembers it when you chat in the portal. * **You can reference past conversations.** "What was that pricing we discussed yesterday?" or "Go back to the proposal we worked on last week" — your agent can recall previous interactions from its memory. *** ## The Thinking Indicator When you ask your agent something complex, you might see a pulsing indicator in the chat. This means your agent is "thinking" — working through the problem step by step before responding. After your agent responds, you can click to **expand the reasoning** and see its thought process: > **Thinking:** The user is asking about a quote for a group of 75. Let me check the pricing guide... The standard rate is $45/person for groups of 50-99. But they mentioned a referral — I should check if the referral discount applies. Yes, 10% referral discount. So: 75 × $45 = $3,375 minus 10% = $3,037.50. I should round to \$3,038 per the rounding rules. This is useful when you want to verify how your agent arrived at an answer, especially for pricing calculations or complex decisions. *** ## Tips for Getting the Best Results You don't need formal language or special syntax. "Hey, can you check if we have a contact for Riverside Insurance?" works perfectly. Instead of sending five short messages with one piece of info each, bundle it together: "New lead: James Park, [james@email.com](mailto:james@email.com), wants the starter plan, was referred by Lisa Chen, expected start date is May 1st." Your agent works better with complete context. "Give me a table," "Keep it to 3 bullet points," "Write it as a formal email," "Give me the short version" — your agent adapts to whatever format you ask for. If you're not sure why your agent made a certain decision, just ask: "Why did you recommend that approach?" or "Walk me through how you calculated that." The fastest way to train your agent is to correct it when it's wrong. Don't just accept an imperfect response — take 30 seconds to say "actually, do it this way instead." Those corrections add up fast. If you'd walk over to a colleague's desk and say "hey, can you look up..." or "can you draft a..." — that's the kind of thing your agent handles well. Think of it as your always-available coworker. "What does our cancellation policy say?" "How many deals do we have in the pipeline?" "What did we decide about the Anderson account?" — there are no dumb questions. Your agent is here to help. *** ## What Happens When Your Agent Is Wrong Sometimes your agent will be wrong, and that's okay. Here's why it happens and what to do: | Why it happened | What to do | | ------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------- | | **Knowledge file is outdated** | Update the knowledge file with the correct information. Your agent will use the new version immediately. | | **No knowledge file exists for that topic** | Upload a document covering that topic. Your agent was guessing before — now it'll have the source. | | **It misunderstood your request** | Rephrase with more context. "That's not what I meant — I wanted X, not Y." | | **It followed the wrong process** | Tell it the right process. "Don't do it that way. Here's how we actually handle it: ..." It'll update its rules. | | **It's a complex calculation** | Double-check the math and tell it if it's wrong. "The total should be $3,200, not $2,800. You forgot to include the setup fee." | The key takeaway: **your agent can only be as good as the information you give it.** The more complete your knowledge files and the more corrections you provide, the fewer mistakes it makes.