TownMaking it yours
Your assistant works well out of the box, and it gets better every day. But there is a difference between an assistant that works and one that feels like yours. This chapter is about closing that gap - refining behavior, correcting what is off, and understanding what is happening under the hood.
The feedback loop
The most important thing to understand about Town is that it is designed to be corrected. Every time you give feedback, your assistant gets better.
Reject a draft and explain why? It adjusts. Tell it a preference? It remembers forever. Tweak a workflow’s instructions? Tomorrow’s run is better.
This is not a one-time setup. It is an ongoing relationship - the same way you would refine how you work with a colleague over weeks and months. The people who get the most out of Town are the ones who treat it this way: not as a tool to configure once, but as a teammate to train over time.
Teaching it your voice
In the early days, your assistant might sound too formal, too casual, or too generic in its drafts. This is normal - it is working from your profile, but it has not had many corrections yet.
You can speed this up. Do not just reject a draft - tell it why. “That is too formal. I would say this more casually, like: ‘Hey Sarah, quick heads up…’” Your assistant stores these corrections and applies them going forward.
You can also go straight to your memory profile and tell it how you communicate: “I write short, casual emails. I open with ‘Hey’ or ‘Hi,’ not ‘Dear.’ I use bullet points and keep things under 5 sentences.” The profile is editable anytime - use it.
Refining workflows
Workflows improve the same way. After a few runs, you will notice things you want to adjust.
Maybe your morning briefing is too long. Say: “Make the briefing shorter - just the top 3 things I need to know today.” Maybe it is including stuff you do not care about. Say: “Stop including calendar events I created for myself - those are just focus blocks.”
Maybe a classifier is mislabeling certain emails. Look at the session history to see why, then sharpen the instructions: “Emails from noreply@ addresses are not always newsletters. Check the content before labeling.”
Each correction is small, but they compound. A workflow that felt 80% right after the first run feels 95% right after a week of small adjustments.
And if a workflow just is not working for you? Turn it off. Every workflow has an enable/disable toggle - there is no penalty for switching one off while you figure out what you want it to do differently. Some people disable a workflow, rethink the instructions, and re-enable it a few days later. That is a perfectly normal part of the process.
One workflow worth special attention: auto-drafts. This is the workflow that drafts email replies for you automatically. Because it touches your voice and your relationships directly, it is the one people have the strongest opinions about. If a draft misses the mark, do not just delete it - tell your assistant what was wrong. “Too formal for this person.” “I would not apologize here.” “Way too long - keep it to two sentences.” That specific feedback is what trains the drafts to sound like you, not like a generic AI. Over time, the drafts get close enough that you are just tweaking a word or two before sending.
When something is not working
Sometimes the issue is not a preference - something is actually off. Here is how to diagnose it.
Start with session history. Every workflow run is logged in full detail - what it read, what it decided, what it did, and what it skipped. This is your diagnostic tool. Instead of guessing, you can see exactly what happened.
You can also just ask your assistant: “Why didn’t my morning briefing include the meeting with Sarah?” It will check the session history and explain.
If a workflow is not running at all, check three things: Is it enabled? Is the trigger configured correctly? Is the right account selected? For scheduled workflows, also check the timezone - a workflow set for 8 AM Eastern runs at 5 AM Pacific.
If a workflow runs but misses things, it is almost always an instruction gap. Your assistant is doing exactly what you told it to - you just told it something slightly different from what you meant. Look for assumptions you made that were not explicit. Did you say “unread emails” but the email was already marked read by another device? Small tweak: “Include emails received in the last 2 hours even if they have been read.”
If your assistant cannot find something, check which account it is searching. With multiple Google accounts connected, it might be looking in the wrong one. For email, check trash and spam - search does not look there by default. For integrations, remember that connecting a tool and enabling it on a workflow are two separate steps (see Chapter 2).
Often, the issue comes back to context. Your assistant is only as good as the information it has access to. If it is missing something, ask yourself: does it have the context it needs to do this well?
Auth and permissions
Sometimes a connected service stops working. You will see errors about authentication or permissions.
The fix is almost always the same: go to Settings, find the affected account, and re-authorize. This refreshes the connection. Think of it like a session expiring - you just need to sign in again.
Restarting a conversation
Long conversations sometimes go sideways. Your assistant might lose track of what you asked or go in circles.
When this happens, start a new conversation with a fresh summary of what you need. For workflows, if the instructions have gotten complicated through many rounds of editing, it can help to rewrite them cleanly from scratch.
Security and privacy
Town is designed with safety and privacy as core principles. You are always in control - the trust dial, per-tool overrides, and the ability to disable any workflow at any time. If something does not feel right, turn it off. You can always dial back.
For the full details on how Town handles your data, see town.com/features/security .
Getting help
Two options:
- Ask your assistant. It can check its own workflow history, diagnose issues, and suggest fixes. Try: “Why didn’t my auto-label workflow process the email from David this morning?” or “My morning briefing was too long today - can you shorten it?”
- Report a bug. Click the bug icon in the bottom-left corner of the web app. It goes directly to the team.
The big picture
Here is what it all comes down to:
Your assistant starts by getting to know you - reading your emails, building your profile, learning your voice. You start by giving it small tasks and checking its work. Over time, the trust builds.
The drafts sound more like you. The workflows get smarter. The suggestions get more relevant.
The things that used to take you 20 minutes happen automatically.
The best way to get there is to keep using it. Start with a conversation. Enable a stock workflow. Build a custom one. Correct it when it is off. Every interaction makes it sharper, and every correction makes it more yours.