Getting Started
What Your Agent Can (and Can't) Do
Set realistic expectations, discover powerful capabilities, and learn how to get the most from your AI agent.
⏱ 4 minute read
Your Agent Is Powerful — But It's Not Magic
Your heyron agent is an AI running on its own private server with its own dedicated workspace. It can do a lot — write code, research topics, manage tasks, chat with you across platforms — but it needs you to set things up and tell it what you want.
Think of it as a brilliant assistant who just started on day one. They're incredibly capable, but they need context, access, and direction from you to do their best work.
What Your Agent Can Do Right Now
✅
Write & Edit
- Draft emails, articles, social posts, essays
- Edit and improve your writing
- Summarize long documents
- Create templates and form letters
✅
Research & Analyze
- Search the web for information
- Analyze data and create summaries
- Compare options and make recommendations
- Fact-check and find sources
✅
Code & Build
- Write Python, JavaScript, HTML, and more
- Build websites, scripts, and automation
- Debug code and explain errors
- Create spreadsheet formulas
✅
Organize & Plan
- Create schedules and to-do lists
- Break big projects into steps
- Set reminders (via cron)
- Track goals and progress
✅
Communicate
- Chat via web dashboard, Discord, Telegram
- Each channel (Discord, Telegram, web chat) has its own separate conversation — they don't share history
- Within a conversation, your agent remembers everything you've said (until the conversation gets very long)
✅
Learn About You
- Stores preferences and context in memory files
- Gets better the more you use it
- Can adapt its personality via
SOUL.md
What Your Agent Can't Do (Yet)
❌
Access Your Accounts (Without Setup)
- Can't read your email, calendar, or social media out of the box
- Needs API keys or credentials you provide
- Each integration requires explicit setup
❌
Take Actions in the Real World
- Can't make purchases or phone calls
- Can't physically control devices
- ✅ But it can draft things for you to copy, send, or post yourself
❌
Remember Everything Automatically
- Has a context window limit
- You need to tell it to save important things — say "remember this" or "write that down." It won't always do it on its own, especially early on
- Won't know what you said last week unless you directed it to save it (or set up rules to save automatically)
❌
Generate Images Natively
- Can write image prompts for DALL-E or Midjourney
- Can create SVG and HTML visuals
- No built-in image generator
❌
Access Other People's Data
- Your agent is private — only knows what you tell it
- Can't access other users' agents or data
- Can search the public web
What People Are Actually Using Their Agents For
Real examples from the heyron community:
- 📊 Building dashboards that display data from their agent
- 📬 Organizing and summarizing their inbox (with email integration)
- 📝 Daily briefings delivered to Notion
- 💰 Tracking futures trading signals
- 🌐 Creating and publishing websites
- 📋 Planning events and managing to-do lists
- ✍️ Co-writing books and content
How to Get More Out of Your Agent
The difference between a "meh" agent and an amazing one usually comes down to a few key steps:
- 🧬 Set up SOUL.md — personality makes everything better
- 💬 Connect to Discord or Telegram — chat from anywhere
- 🧠 Understand memory — use it effectively
- 🔍 Ask your agent what it can do! — literally say:
What tools and skills do you have?
- 📚 Browse the Agent Library — community skills and templates
🌱 It grows with you
Your agent gets better the more you use it. The people who love their agents most are the ones who invested time in SOUL.md, built up memory, and experimented. Start with one task and grow from there.
Quick Reality Checks (from recent community questions)
- 🌐 Browser expectations: your agent can search/fetch/read web pages, but it generally cannot click through JavaScript-heavy flows or fill forms like a human browser session.
- 📣 Social media + website expectations: your agent can write posts, content calendars, captions, and website code/content drafts — but direct publishing and final launch steps usually still need your account permissions and a human click to complete.
- 📌 Discord action limits: your agent can read, draft, and reply, but some native moderation UI actions (like pinning messages in-channel) may still be unavailable even when Discord permissions look correct.
- 💸 Token usage anxiety: heavy model usage can consume plan/budget quickly. Ask for a token-aware mode (short answers, fewer retries, smaller context) and check your billing/plan reset window before assuming the system is broken.
- 🎙️ Voice setup gotcha: if ElevenLabs voice is connected but speech still fails, verify you selected a voice that exists in your own ElevenLabs account and shared the correct voice ID + API key with your agent.
- 🕒 Time-of-day weirdness: if your agent says “good night” at 3pm, ask it to check current local time and timezone, then save your preferred timezone in memory.
- 🌙 "Work while I sleep" expectation: closing a chat tab can stop in-thread work. For guaranteed background runs, use explicit scheduling (cron/reminders/heartbeat) and ask for a completion message in your target channel.
- 🧵 One agent, many surfaces: Discord DM and server channels are the same agent identity, but each surface has separate conversation context/history.
Automation Reality Check: Popular Requests
- 🎨 Canva: your agent can usually help with copy, layout direction, and design prompts. Direct in-app editing/publishing depends on connector availability and permission scope, so verify with a tiny canary task first.
- 🖼️ Midjourney: your agent can generate prompt packs and workflow steps. Direct control is often constrained by Discord/bot interaction limits, so plan for a manual final trigger unless a proven integration is confirmed.
- 📱 Instagram/TikTok: content planning and caption generation work great; direct posting is commonly blocked by account policy/API constraints. Use a “final-post handoff” checklist to keep output shipping.
- 🌍 Website builds: your agent can produce real HTML/CSS/JS and deployment steps, but DNS/account-level publish actions may still require your credentials and one final click.
Use a 2-minute canary before long tasks
Before a big workflow, ask for one tiny proof action (for example: one Canva draft change, one test social post draft, one single-file website publish plan with exact destination). If the canary fails, switch to handoff mode immediately instead of burning tokens on retries.
Fast self-check prompt
Paste this to your agent: List my current tools/integrations, my saved timezone, whether you can browse/click websites, whether you can directly publish to social platforms from my current setup, whether you can pin messages in Discord from this environment, and whether you can keep working after I close chat. Also switch to token-aware mode (concise responses, minimal retries) and show me where to confirm my plan/reset limits. Then run a 2-minute canary for my requested workflow and tell me if we should use direct automation or manual handoff.
What's Next?