Withmere
I’m building Withmere, a personal AI environment for life and work. Through Discord and Telegram, I talk with agents that have distinct roles, think through decisions and work on projects. The system connects those conversations to tools, project workspaces, memory and scheduled work.
Withmere is in private development and personal use. The code and binaries are not publicly distributed. Here I share selected accounts of building and using the system.
Read the development account →A development account
What it takes to add a feature to an existing system
I wanted an agent to be able to start a conversation. The feature worked, including accepting my reply. Later changes exposed disagreements about how its messages were stored and which conversation a reply belonged to.
- The Note follows that history through product requirements, shared conversation rules and tests.
- It draws on private development records and reports application-level regression checks; it does not establish recovery of the running service.
- The case shows why maintaining an ongoing interaction requires work beyond generating and delivering its first message.
Picking up where I left off
An ongoing question rarely fits inside one conversation. I might need to think it through, work on it, leave it for a while and return with a different understanding. I want the useful context to remain available without having to explain everything again.
That includes ordinary conversation and companionship as well as practical work. Something that matters may need an action, another conversation, or a deliberate decision to leave it alone. The product I want to build makes room for each of those outcomes.
From a conversation to work on a project
The messaging apps are the everyday entrance. Behind them, Withmere brings together an agent’s role, selected context and tools, and the model backend that executes the conversation. I can select a project explicitly; a compatible execution backend then works in that project’s directory.
The system also manages conversation continuation, stored history and timed work. These mechanisms have to agree when an existing conversation ends, a new one begins, or a scheduled action reaches me. A message arriving in a chat is only one part of that behavior.
Keeping memory and action accountable
The design separates records of what happened from the lasting picture the system builds of me. Memory recommendations have review and application steps, so accepting a suggestion and changing a stored fact remain distinct events. I want to be able to examine and correct what the system carries forward.
There is a similar distinction in following through. Delivering a reminder does not establish that I did what it reminded me to do. For work executed through the system, its actions and results can leave records to inspect. For life outside it, my account of what happened still matters.
What I want to learn next
I intend to develop Withmere into a service. The next product question is which parts of this environment would help someone else, and what they should be able to use without maintaining the system as I do. The delivery model and pricing are still open.
Background and evidence
Withmere is the product name for work developed privately as AgentOS. The project description draws on its product documents, architecture and implementation; the linked Note traces a specific development history. The underlying repository and conversation logs remain private. This page does not offer a live demo or independently reproducible implementation, and it does not claim outside adoption or uninterrupted reliability.
- Type
- A personal AI environment for life and work
- Domains
- personal AI / human-agent collaboration / conversation continuity / memory / project execution
Available material
- A written development account of proactive conversations, their later integration failures and reported regression checks, based on private records.