Semuth
Semuth is an AI workspace for software engineering. It brings coding agents, models with different roles, and team workflows into one place.
The idea
I’m building Semuth to explore a better way to use AI in day-to-day software engineering, based on how I work on production software myself.
The idea is that different kinds of work suit different models, and that agents should be able to act on a codebase, not just talk about it.
The models
Black
Black — for fast, everyday work.
Weaver
Weaver — for deeper work that benefits from more reasoning. Its effort level can be adjusted to the task.
Current development
Model roles — Black and Weaver
Black for fast, everyday work; Weaver for deeper problems.
Team workspace
A shared workspace for teams, with member management and shared credits or packages.
Experiments
Coding agents
Coding agents that can work with your codebase — finding relevant files and context, proposing changes, and checking their output. Still experimental.
Planned directions
Remote collaboration
Remote CLI connections and shared workflows, with room later for other roles like QA and PM.
Research direction
Model research
Longer-term research includes retrieval, model adaptation, self-hosted inference, and deeper model work. These are research directions, not current capabilities.
Where it is going
Casual chat and image generation are also being considered.
Software engineering stays the focus — that’s what Semuth is for.