> ## 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.
# How Your Agent Works
> Everything you need to know about your Spark agent — what it is, how it thinks, and what it can do for you
So you've signed up for Spark and you're looking at a chat panel thinking, *"What do I actually do with this thing?"*
This page will answer that question — completely. By the end, you'll understand exactly what your agent is, why it's useful, and how to start using it right now.
***
## First Things First: What Is an Agent?
Imagine hiring a new employee. On their first day, you'd tell them about your business — your services, your pricing, how you like things done. Over the next few weeks, they'd learn your processes, remember your clients, and eventually handle things without you having to explain every detail.
**Your Spark agent is that employee — except it works 24 hours a day, never forgets anything, and gets up to speed in days instead of months.**
It's not a chatbot that gives you the same canned answers over and over. It's not a search engine. It's an intelligent assistant that:
* **Learns your business** because you teach it — just like you'd train a new hire
* **Remembers everything** you tell it — client names, your preferences, rules you've set, corrections you've made
* **Takes real action** — it doesn't just answer questions, it actually *does things*: creates contacts, writes emails, generates reports, processes leads, updates your pipeline
* **Works while you sleep** — it runs 24/7, handling scheduled tasks and incoming leads even when you're not at your desk
**You cannot break your agent.** This is important. There's no wrong thing to type, no button that will mess everything up. Talk to your agent like you'd talk to a coworker. If it does something wrong, just tell it. It'll fix it and learn from the mistake.
***
## What Can Your Agent Actually Do?
Let's make this real. Here are things you can do with your agent *right now*, with the exact messages you'd type:
### Manage Your Customers
Instead of manually typing customer information into a spreadsheet or clicking through screens to create records, just tell your agent:
> **You:** I just got off the phone with Maria Garcia. She's interested in our premium package. Her email is [maria@example.com](mailto:maria@example.com) and her number is 555-0142.
>
> **Agent:** Done! I've created a contact for Maria Garcia with her email and phone number. I've added her to the "Interested" stage in your pipeline and noted her interest in the premium package. Want me to draft a follow-up email to her?
One message. Contact created, pipeline updated, notes added, and it's already offering to draft the follow-up. That's what would have taken you 5 minutes of clicking through screens.
### Get Answers About Your Business
Once you've uploaded your pricing guides and policies (we'll cover that in [Knowledge Files](/agent/knowledge)), your agent becomes an instant expert:
> **You:** A client is asking about our cancellation policy. What do I tell them?
>
> **Agent:** Based on your policy document, here's what to tell them: Cancellations made 48+ hours in advance receive a full refund. Cancellations within 24-48 hours receive a 50% credit toward a future booking. Same-day cancellations are non-refundable. Want me to draft a response email with this information?
No digging through folders or flipping through binders. Your agent checked the right document and gave you the answer in seconds.
### Create Documents and Reports
> **You:** Write me a proposal for the Anderson account. They want our gold package for a team of 25, starting in April.
>
> **Agent:** I've created a proposal for the Anderson account. It includes the gold package pricing for 25 users (\$2,500/month), the April start date, our standard onboarding timeline, and the terms from our template. You can review it in your Work tab. Want me to adjust anything?
Your agent wrote a full proposal using your pricing guide and template — ready for you to review, edit, and send.
### Handle Leads While You Sleep
This is the one that makes people go *"oh, THAT'S why this is powerful."*
You set up a connection between your website contact form and your agent (we'll show you how in the [Lead Capture guide](/guides/lead-capture)). Then:
**11:47 PM** — Someone fills out your contact form. You're asleep.
**11:47 PM** — Your agent instantly processes the submission, creates a contact record, calculates a quote based on what they asked for, drafts a personalized response email, and saves everything.
**7:00 AM** — You wake up, open Spark, and see:
> **Agent:** While you were away, I processed 3 new leads:
>
> 1. **James Rivera** — Interested in the starter package for a team of 10. Estimated quote: \$750/mo. Draft response ready.
> 2. **Priya Patel** — Asked about custom enterprise pricing for 200+ users. Flagged for your review since it's above the standard tier.
> 3. **Tom Baker** — Requested a callback. Added to your task list for today.
>
> Want me to send the draft responses, or would you like to review them first?
That's three leads handled before your morning coffee. Without your agent, those leads would be sitting in an inbox waiting for you to manually respond.
### Run Things on Autopilot
> **You:** Every Monday morning at 8am, I want you to check the pipeline, see which deals haven't had activity in the past week, and send me a summary of who I need to follow up with.
>
> **Agent:** Got it! I've set up a weekly schedule: every Monday at 8:00 AM, I'll review your pipeline for stale deals and send you a follow-up summary. You'll see it here in chat when it's ready. Want me to also check for any leads that haven't been contacted yet?
Now every Monday, without you lifting a finger, you get a ready-made follow-up list.
### And Much More
"Check my email for anything urgent" · "Schedule a call with the Peterson team for Thursday at 2pm" · "Draft a reply to that email from David"
"Look up recent news about renewable energy incentives in Arizona" · "Find me three competitors in the Portland area and summarize their pricing"
"Build me a dashboard showing new leads this week, total pipeline value, and our conversion rate" · "How many deals did we close this month?"
"Remind me to call Sarah on Thursday at 2pm" · "What's on my task list for today?" · "Create a task for the team to review the Q2 proposal by Friday"
***
## How Your Agent Thinks (Behind the Scenes)
You don't need to understand this to use your agent — but knowing how it works helps you get better results.
When you send a message, here's what your agent does in the background:
Your agent reads its personality file ("I'm Alex, a friendly sales coordinator who's always professional but approachable") and its instruction file ("When a lead asks about pricing, always check the pricing guide first"). This happens every time, which is why your agent is always consistent.
Your agent reads its notes from today and yesterday, plus its long-term memory. This is how it remembers that you told it last week that the Johnson deal is high priority, or that you prefer bullet points over paragraphs.
Your agent processes your message in the context of everything it knows — your conversation history, your business rules, your preferences, and any relevant background.
If your question touches on something in your knowledge files — pricing, policies, processes — your agent pulls up the relevant document and reads it. It doesn't guess from memory; it checks the source.
Based on all of the above, your agent decides what to do. It might answer your question, create a record in your CRM, write a document, run a search, or some combination. Then it responds.
If something important happened — a new client preference, a correction you made, a decision — your agent notes it in its memory so it doesn't forget.
When your agent is working through a complex request, you'll see a **thinking indicator** in the chat — a small animation that means it's reasoning through the problem step by step. After it responds, you can expand the reasoning section to see exactly how it arrived at its answer. This is useful when you want to understand *why* it made a certain decision.
***
## The Four Things That Shape Your Agent
Your agent's behavior comes from four sources. Think of them as layers that work together:
### 1. Personality — Who Your Agent Is
This is your agent's identity. It controls things like:
* What your agent calls itself ("Hi, I'm Alex!")
* How it communicates — formal or casual, brief or detailed, serious or friendly
* What it considers important — "Always follow up promptly" or "Never make promises I can't keep"
**You set this up when you create your agent**, but you can change it anytime just by asking:
> **You:** From now on, keep your responses shorter. I like bullet points, not long paragraphs.
>
> **Agent:** Noted! I've updated my communication style. I'll keep things concise and use bullet points going forward.
The personality is like your agent's *character* — it stays consistent across every conversation, every channel, every task.
### 2. Instructions — The Rules Your Agent Follows
Instructions are your agent's playbook. They tell it how to handle specific situations:
* "When calculating quotes, always round up to the next package size"
* "When a lead mentions they were referred by an existing client, give them the referral discount"
* "Always check the calendar before suggesting meeting times"
**You build these up over time.** Every time you correct your agent or give it a new rule, it adds to its instructions:
> **You:** Actually, we stopped offering the 20% early-bird discount. Remove that from your rules.
>
> **Agent:** Updated — I've removed the early-bird discount from my pricing rules. I'll use the standard pricing from now on.
### 3. Knowledge — Your Business Documents
Knowledge files are documents you upload that teach your agent about your business:
* Pricing guides and rate sheets
* Company policies (returns, cancellations, warranties)
* FAQs and common questions
* Product descriptions and service details
* Email templates and response scripts
* Team info, org charts, anything a new employee would need
Your agent doesn't memorize these — it's more like having a reference binder on its desk. When a relevant question comes up, it reaches for the right document and reads it.
**The huge advantage:** You can update a knowledge file anytime, and your agent sees the new version immediately. Changed your pricing? Update the document, and every answer from that point forward uses the new numbers.
### 4. Memory — What Your Agent Remembers
Your agent has three layers of memory:
| Layer | What it is | How long it lasts |
| -------------------- | ------------------------------------------------------------------------------------- | ----------------------------- |
| **Conversation** | What you're currently talking about | Until the conversation ends |
| **Daily notes** | A daily journal of what happened — key decisions, corrections, important interactions | Permanent — one file per day |
| **Long-term memory** | The most important facts, curated and organized by your agent over time | Permanent — updated regularly |
This is why your agent can pick up where you left off, even if you haven't talked to it in days. It reads its recent notes every time it starts working.
> **You:** Remember that the Henderson project is on hold until they get board approval. Check back with them in June.
>
> **Agent:** Got it — I've noted that the Henderson project is on hold pending board approval, and I'll flag it for a June follow-up.
Six weeks later:
> **Agent:** (Monday morning report) Quick reminder: you asked me to follow up on the Henderson project in June — their board approval was pending. Want me to draft a check-in email?
***
## How Your Agent Gets Better Over Time
This is the part that surprises most people. Your agent isn't a static tool — it genuinely improves the more you use it.
### Week 1: The Training Phase
Your agent is like a new employee on day one. It's smart and capable, but it doesn't know your specific business yet. You'll spend time:
* Uploading your key documents (pricing, policies, FAQs)
* Explaining your processes ("Here's how we handle new leads...")
* Correcting mistakes ("No, that's not how we calculate shipping")
* Setting preferences ("I like it when you format reports this way")
**This is normal and expected.** Every correction makes your agent smarter.
### Weeks 2-3: The Ramp-Up
Your agent now knows your most common workflows. It handles routine tasks without detailed instructions. You're correcting less and delegating more:
> **You:** New lead from the website — process it.
>
> **Agent:** Processed! I've created a contact for Lisa Chen, calculated a quote for the standard package (\$1,200), drafted a welcome email using our template, and added her to the pipeline at the "New Lead" stage. The draft email is ready for your review.
Notice you didn't have to spell out every step — your agent already knows your process.
### Month 2 and Beyond: The Payoff
Your agent feels like a team member who's been with you for years. It knows your clients, your preferences, your processes. It anticipates what you'll need:
> **Agent:** (Monday morning) Good morning! Three things for today:
>
> 1. The Rivera proposal expires tomorrow — want me to send a reminder?
> 2. You have a call with the Whitfield team at 2pm. Based on our last conversation, they're likely going to ask about the enterprise tier — I've pulled the pricing sheet for reference.
> 3. Two new leads came in over the weekend. Both are standard — I've processed them and draft responses are ready.
That's an agent that *knows* your business. And it got there because you invested the time to teach it.
***
## Always On, Always Available
Your agent isn't something you "open" and "close." It's running 24/7:
* **Processes leads at midnight** when someone fills out your contact form from the other side of the world
* **Runs scheduled reports** at 6am so they're waiting for you when you wake up
* **Responds instantly** when you send a message — no loading time, no waiting
* **Remembers everything** even if you close your browser and come back a week later
* **Picks up where you left off** because it reads its memory at the start of every session
***
## Multiple Ways to Reach Your Agent
You're not limited to the Spark portal. Your agent can connect to the messaging apps you already use:
* 💬 **Spark Portal** — the web chat panel in your workspace
* ✈️ **Telegram** — chat with your agent from your phone
* 💼 **Slack** — bring your agent into your team workspace
* 🟣 **Discord** — add your agent to your server
* 📱 **WhatsApp, Signal, SMS** — and more
It's the **same agent** on every channel — same knowledge, same memory, same personality. You can ask it something from your phone while walking the dog, then continue a different conversation from your computer later.
Learn more in [Channels](/agent/channels).
***
## Where to Go Next
You understand how your agent works. Now it's time to start using it:
Learn how to talk to your agent effectively, with dozens of example prompts you can copy and use right now
Upload documents that turn your agent into an expert on your services, pricing, and policies
Step-by-step guide to teaching your agent your business workflows through simple conversation
Discover why you barely need to touch the interface — your agent can handle almost everything for you