Private AI Agent
A personal AI agent that runs a client's daily routines from Telegram and Slack, from calendar and inbox to ad checks across Google and Meta.
Private client · 4 weeks, brainstorm to handover
- Client
- Private client
- Sector
- AI Agents
- Team
- 3-person team
- Timeframe
- 4 weeks, brainstorm to handover
The problem
The client's day was spread across Gmail, Calendar, Drive, Google Ads, Meta Ads, WhatsApp, Instagram, and two chat apps. Keeping up meant checking each one by hand, several times a day. They wanted one assistant they could simply message, and one that handled the recurring checks without being asked.
Constraints
It had to live where the client already talks, in Telegram and Slack, not in a new app. It needed access to Google and Meta accounts without ever acting in the client's name unchecked. Running costs had to be known before anything was built. And the client had to understand how every part worked, not receive a black box.
What Trace built
A private AI agent with four parts behind one chat. A model reads each request and plans the steps. Memory, stored in PostgreSQL, keeps preferences, people, and past requests. Routines run recurring checks on a schedule: a morning brief from the calendar, inbox, ads, and customer chats, ad spend and cost-per-lead checks through the day, and weekly numbers filled into Sheets. Connections reach the client's Google and Meta accounts, and guardrails keep a human in the loop: nothing is sent, posted, spent, or deleted until the client taps yes. The engagement ran from brainstorm and budget through the build to a part-by-part walkthrough.
Team
Timeframe
4 weeks, brainstorm to handover
Outcome
The client now runs their routine from one chat. Recurring checks happen on schedule, and the agent stays human-in-the-loop: nothing goes out in their name, and no budget changes, without a yes. Costs were agreed before the build started, and at handover the client knew what each part does and how to change it.
Questions
About this project.
What did Trace build?
A private AI agent the client reaches in Telegram and Slack. It connects to their Google and Meta accounts, remembers context, runs scheduled routines, and asks before anything goes out in their name.
How was the project run?
A 3-person team took it from brainstorm and budget through the build, then walked the client through each part before handover.
Does the agent act without permission?
No. It works human-in-the-loop. Anything it would send, post, spend, or delete waits for the client's approval, and it can only reach the accounts the client connects.
Are the chats on this page real?
No. They are illustrative examples of the routines the agent runs. The client's real chats and accounts stay private.
Next case study
Your project
Building something with similar stakes? Tell us about it. We'll say plainly whether we're the right team.
Book a call