About

The story

A setup, not a framework. Built for one operator, kept honest by daily use.

Why it exists

Most assistant setups fail the same way. They forget everything between sessions, they pile up brittle prompts with no operating system around them, or they automate isolated tricks instead of building a durable loop.

Clawrari is the opposite bet. It treats the assistant like infrastructure: file-first memory as the system of record, git as the audit log, explicit rules for startup and approvals, and review loops that catch drift before it ships. The point is not a single clever prompt. It is a small set of reliable systems that compound.

The two-repo model

Clawrari is split on purpose.

clawrari is the setup itself: skills, memory architecture, model playbooks, harness, and crons. Clone it and bootstrap a fresh install.
clawrari-context is how it gets built: the FRDs, blueprints, and process behind the decisions. The product-development context, open-sourced.

The setup shows you the what. The context repo shows you the how and the why. Read both: clawrari and clawrari-context.

Who builds it

Prasith runs Clawrari as his own daily driver. It is not a side demo; it is the system he uses to do real work, which is why the choices are opinionated and the numbers are verifiable. More at prasithg.github.io.

Honest scope

This is one operator's production setup, opinionated by design. It is not a neutral starter template and it does not try to please everyone. Some choices will not fit your workflow, and that is fine. Take the patterns that hold up, leave the rest, and keep what compounds.

View on GitHub Back home