Personal AI for a private client
I built and deployed a personal AI system for a private client, connecting a Telegram assistant, specialist agents and an Obsidian workspace through OpenClaw. The engagement covered the system setup, onboarding, workflows and operating instructions that made those pieces useful together.
A place for the work to live
The system supported recurring briefings, research, inbox routing and shared deliverables. Telegram gave the client a familiar place to make requests, while the workspace held the notes and outputs they could return to. I organized the work so a conversation could lead to something useful that remained available afterwards.
Behind that interface, I worked on where information belonged, how ongoing tasks were recorded and how the assistant recovered context to continue them. Specialist agents needed clear responsibilities, and their results needed to come back into a coherent experience for the client.
Knowing when to act
I shaped the rules for when the assistant could act and when it needed a human decision. Sensitive workflows had separate responsibilities and explicit review points. Onboarding and instructions made those expectations part of how the system was used.
Useful tools need clear boundaries.What the deployment taught me
The work made the relationship between memory, useful outputs and day-to-day interaction concrete. It also showed how much of the experience depends on the connections between tools: remembering enough to continue, making progress visible and giving a person a clear place to find the result.
Those lessons now inform Enlo, the broader personal-AI product I'm developing. This deployment was a bespoke system built around one client's needs, with its own scope and delivery.