About the lab
An assistant should earn its place at home.
Sharper Than AI is an independent build notebook about an always-on personal AI assistant running from a home server. The server owns context, permissions, memory, schedules, integrations, and logs. Each workflow can use rules, a small local model, or an explicitly approved cloud model.
The question behind the project
A storage-focused home server is not automatically a good replacement for a GPU workstation or a hosted frontier model. It can still be the best place to run an assistant: it is always available, close to household data and services, and capable of coordinating reliable automations.
The project asks which parts genuinely benefit from staying local, when cloud inference is the proportionate choice, and whether the completed workflow saves more effort than it creates.
What counts as evidence here
Architecture notes and build plans are labelled as such. A benchmark needs the hardware, software versions, model, quantization, context, prompt set, and measurement method. A successful automation needs more than one demo run: it needs visible failures, corrections, latency, cost, and maintenance.
When a post relies on external technical behavior, it links to primary documentation. When results are not measured yet, the article says so instead of turning an expectation into a conclusion.
The secondary project
Sharper also maintains an open coding-agent instruction evaluator. It prepares and runs controlled comparisons for AGENTS.md and related repository instructions. That work is kept as a distinct project because it already has a working tool, local runner, example data, and a reproducible methodology.
Read how the local evaluation runner works or inspect the source repository on GitHub.
Follow the build, including the misses.
The lab notes cover architecture, measured experiments, failed ideas, and the maintenance work that usually disappears from an AI demo.