Training SummaryOverview
This session introduced the team to your personal AI agents running on Discord, the dashboard system hosted on solonox.com, and the roadmap for automating email, HubSpot workflows, and employee monitoring. By the end of this guide, you should be able to chat with your agent, start building your first dashboard, manage your context window, and begin setting up Google Workspace API access.
One thing at a time. If you give your agent 30 things to do at once, it will mess up every single one. Give it one task, test it, confirm it works, then move to the next. This applies to dashboards, API setup, email automation — everything.
First StepsGetting Started with Your AI Agent
The original onejan.com links from earlier emails no longer work. Peter moved everything to solonox.com to prevent the AI server from interfering with the company website. All dashboard and SOP links now use the solonox.com domain (e.g., bco-board.solonox.com, sms-plan.solonox.com). Use the links spreadsheet for the correct URLs.
Core WorkflowBuilding Your Dashboard
Your first real task is building out your department dashboard. Peter wants you to start here because you can't break anything with a dashboard — it only reads data, it doesn't change anything in HubSpot. This is your safe playground to learn how the agents work.
Taylor (Client Services): Tab 1 = CS team (Jean, Dave, Paul). Tab 2 = Recruiter. Start with: calls per person, emails sent, tickets opened today, tickets open longer than 7 days.
Lobsang (Sales): Tab 1 = Appointment Setters. Tab 2 = Presenters. Start with: sat rate, calls made, talk time, appointments scheduled, schedule quality (with vs. without commitment).
Farouk (BCO): Already partially built. Tabs for BCO Scout, BCO Filler, BCO Setup. Shows: stage movements, total calls, emails, call time, downtime (gaps over 5 minutes between calls).
You can open your dashboard URL, see live numbers for your team, hit refresh throughout the day and see updates, and go back to previous days to see historical snapshots (snapshots are taken at ~11:30 PM nightly).
If your team isn't tagging properties correctly in HubSpot, the dashboard will show bad data. The AI can only pull what's there. This is the foundation — if the data input is wrong, everything downstream is wrong. Make sure your people are labeling tickets, statuses, and properties correctly before you rely on dashboard numbers.
Critical KnowledgeContext Window & Resetting Your Agent
Your agent has a memory limit called a "context window." Think of it like a notepad that can only hold so many pages. Once it fills up (around 50-60%), the agent starts forgetting things and making mistakes. Here's how to manage it.
/reset clears YOUR agent's context window. /restart restarts the entire program for EVERYBODY. Do not type /restart. Ever. It kills all agents for the whole team.
Next StepSetting Up Google Workspace API Access
Once you're comfortable with your dashboard, the next step is giving your agent access to your Google Workspace (Gmail, Drive, Calendar, Docs, Sheets). This is what enables it to read and send emails, manage your calendar, and work with your files.
Your agents come pre-loaded with HubSpot API access. You don't need to set this up. If your agent says it can't access HubSpot, tell it: "Check your memory — you should have our HubSpot API key already." Sometimes the agent needs to be reminded to look at its stored memory.
Once set up, you can say things like: "Send Clarissa an email saying [X] — post it in the chat for me to scan first, then send it." Or: "Check my email every 15 minutes for a response from [person]." Or: "What emails do I have to respond to? What's on my calendar today?"
AI API usage is expensive right now. Peter burned ~$700 in one day between setting up the server and rebuilding after a crash. Use your agents with purpose — don't loop them on tasks unnecessarily or leave them running checks you don't need. Be intentional with every request.
From the SessionQuestions & Answers
Q: The training document mentioned chatting with the AI bot on Telegram. Do I not have access to that yet?
A: Forget Telegram for now — you'll be chatting with your agent on Discord. The team moved to Discord because everyone has company emails set up there. Note: Discord runs about 30% slower than Telegram. If it becomes unbearably slow, Peter may switch the team to Telegram later.
Q: Is "Alliances Available" the number of partners we currently have in buildings?
A: No. "Alliances Available" counts only current partners who are waiting for buildings OR whose scouting status is active. It specifically calculates people who want contracts and are eligible — not everyone in the system. This prevents the numbers from being skewed by partners who are happy with their current buildings.
Q: When I'm building my dashboard, should I ask the Sales Agent or the AI Team channel?
A: Talk to your own agent. Sales Agent is Lobsang's. BCO Agent is Farouk's. CS Agent is Taylor's. Each person uses their assigned agent for everything — dashboards, questions, tasks. The team channel (One Gen Team) is for cross-bot communication only.
Q: You said reset and then restart — which one is it?
A: It's /reset. Always /reset. /restart reboots the entire program for all agents across the whole team. Peter has made this mistake himself — it's a painful one. If you need to stop something mid-build, type abort instead.
Q: Is the Google Workspace API setup also integrated with HubSpot, or only personal Google?
A: They're separate. Your agents already come with HubSpot access built in — you don't need to set that up. The Google Workspace API setup is for Gmail, Drive, Calendar, Docs, and Sheets. If HubSpot ever seems disconnected, tell the agent to check its memory for the API key.
Q: When the chat has been going too long and starts breaking, how will I know? Will it say something?
A: You can ask it directly: "What's your context window at?" It'll give you a percentage. You can also tell it: "Let me know when you're at 50%." But if the window gets high, it may forget you asked. Signs to watch for: slow responses, not following instructions, repeating itself, or producing output that doesn't match what you asked for.
Q: What should be the first thing I trigger for my client dashboard?
A: Start super basic. For each CS rep (Jean, Dave, Paul): how many calls today, how many emails sent, how many tickets came in today, how many tickets have been open longer than 7 days. Then add a second view for critical items: tickets with cleaning more than once a week that haven't had a response yet. Don't jump to complex — get the basics working first, then layer on.
Q: When I set up dashboard triggers, does everything in HubSpot need to be labeled properly for it to work?
A: Yes — 100%. If your team isn't labeling tickets, statuses, and properties correctly in HubSpot, the dashboard will show garbage. The AI can only read what's there. The good news: once dashboards are working, the next step is building agents that flag when properties haven't been updated (e.g., "30 of his 100 calls today had no property change — find out why").
Q: Can I make changes to the BCO dashboard as I see fit?
A: Yes — it's yours. Peter specifically stopped building dashboards for managers because he was building them based on what he saw as important, not what the manager sees as critical. You own your dashboard. Make it work for how you actually manage.
Non-NegotiablesKey Rules & Standards
- ✓One task at a time. Give the agent one instruction, test it, confirm it works, then move on. Batch requests = broken output.
- ✓/reset — never /restart. /reset clears your agent's context. /restart kills everyone's agents.
- ✓Always back up before resetting. Ask for a summary, then have the agent save it to your Department Agent Restart file.
- ✓Tell the agent to read its memory. Agents sometimes forget stored information. Say "check your memory" when it claims it can't do something you know it can.
- ✓All links are on solonox.com now. Nothing lives on onejan.com anymore. Use the shared "One Janitorial Links" spreadsheet for correct URLs.
- ✓Use "abort" to stop a build mid-process. If the agent is going off track, type "abort" and then redirect: "Let's go back to step one. Just do this right now."
- ✓Demand iterations before accepting "done." When building something, tell the agent: "I want three iterations before you tell me it's done. Iteration 1: read all the code. Iteration 2: read again and compare to the website. Iteration 3: find the bug if it's not showing up."
- ✓Don't spend all day on this. It's addictive. Play with it, learn it, but keep doing your actual job. Come back to Peter end of week with notes and questions.
- ✓Ask about API rate limits. Before the agent runs bulk operations on HubSpot, ask: "Is this going to hit my API rate limit?" HubSpot limits how fast you can pull data — operations need to happen in sequence, not all at once.
- ✓5 minutes = generous downtime. On the BCO dashboard, downtime is measured as any gap over 5 minutes between calls. Lunch and breaks are already subtracted. Anything showing up as downtime is real idle time.
Pro MovesTips & Shortcuts
Instead of reading long documents yourself, paste the link to your agent and say: "Can you read this whole thing and show me how to do this?" It will break it down step by step in simpler language. This works for any document, process guide, or SOP.
Ask your agent to analyze how the BCO board on bco-board.solonox.com was built — what properties it pulls, how the data maps, what the structure looks like. Then use that as a blueprint for your own dashboard so you're not starting from zero.
Peter had the AI pull population data for cities from Google and add it as a column in the dashboard. Your data sources don't have to be limited to HubSpot. If you need external data (market stats, population, distances), just ask the agent to pull it.
If your agent says it can't do something, try: "How have you done it before for other people?" or "How could we do it? What if we changed this?" Peter's experience: pushing back often triggers the agent to find a solution it didn't initially offer.
Your bot can talk to Peter's bot (Solonox). In the One Gen Team channel, you can tag @Solonox to ask questions about server setup, how things were built, or to troubleshoot. Your bots can collaborate.
Peter's direct quote: "The more I use it, the more I understand I have s***** communication — because this thing does what you tell it to do. If it's not doing what you told it to do, it's because you're telling it wrong." If output is wrong, re-examine how you asked for it.
Additional RecommendationsClaude's Suggestions
These are Claude's recommendations based on the training content — not Peter's instructions. Use your judgment on whether to apply them.
After every /reset, you'll need to re-orient your agent. Write a standard opening prompt that you copy-paste each time: who you are, what department you manage, what you're currently working on, and a pointer to your restart file. This saves 5-10 minutes of re-explaining every reset.
When something goes wrong with your agent (and it will), write down what you asked, what happened, and how you fixed it. This becomes your personal playbook. Within two weeks, you'll have a library of solutions for the common problems and you'll stop making the same mistakes.
Before trusting any dashboard metric, spot-check it. If the dashboard says Paul made 12 calls today, go into HubSpot and count. Do this for your first 3-5 metrics. Once you trust the data pipeline, you can stop manually verifying. But don't skip this step — catching a bad property mapping early saves you from building an entire dashboard on bad data.
Before building anything, ask: "Before you do this, walk me through the exact steps you'll take, the exact properties you'll pull, and where the data comes from. Don't execute yet." Review the plan. Then say go. This is Peter's exact advice applied as a habit — it prevents the agent from guessing wrong and building something you have to tear down.