skaft

skaft

we have AGI at home

Local-first AI tools — so the means of making don't belong to the labs.

↓ the toolkit
// the toolkit

Three tools. One workshop.

relay

the gateway

One endpoint in front of any model — local llama.cpp servers or cloud APIs (OpenAI, Anthropic, DeepSeek). Auto model lifecycle, OpenAI/Anthropic-compatible, observability built in.

curl -fsSL https://raw.githubusercontent.com/achuthanmukundan00/relay/main/scripts/install.sh | bash
GitHub docs

hamr

the agent

A local-first coding agent built for consumer GPUs. Relay-first, OpenAI-compatible, with an embeddable SDK and a fast differential-rendering TUI. Built for your machine, not a cloud bill.

npm install -g @skaft/hamr
GitHub npm docs

suitcase

the copilot

A private AI career copilot. Bring your own LLM; your data stays on your machine. Builds your profile, scores and tracks job leads, tailors materials — and never auto-submits anything.

curl -fsSL https://raw.githubusercontent.com/achuthanmukundan00/Suitcase/main/scripts/install.sh | bash
GitHub docs
// mission

The oldest lesson still holds: whoever owns the picks and shovels owns the dig. Bubble or not, we refuse a world where a handful of labs hold the only tools that actually work.

skaft forges local-first AI infrastructure — sharp, self-hosted, and yours. Built for the garage, the dorm, the workshop bench. Not a platform you rent. A set of tools you keep.

Quality metal outlasts the winter. So do good tools.