v0.5.0 · open source · MIT

clawrari

The Ferrari of OpenClaw setups.

An opinionated, battle-tested OpenClaw setup: a memory architecture, meta-learning loops, a content engine, and agent-orchestration playbooks. One operator's production system, open-sourced.

View on GitHub Quick start

Skills 8 Docs 44 CI selftests 4 Live CI ↗

Philosophy

Five things it believes

Opinionated defaults. Not 47 options. One setup, with a reason for every choice.
Battle-tested. Everything here runs in production. If it is in Clawrari, it is in use.
Your AI should have a personality. Not a corporate drone, not a sycophant. A sharp partner with opinions.
Memory is the foundation. An assistant that forgets everything between sessions is just a search box.
Errors are bugs. A failed cron is not a footnote. It is something to fix now.

What's inside

The map

Four layers, wired together. Read the source; nothing is hidden state.

/skills

Skills

Reusable workflows for coding, review, night work, publishing, and loop-scouting.

/reference

Playbooks

Model routing, prompt guides, agent templates, and the effort ladder that decides what runs where.

/docs

Memory + harness

File-first memory, the dreaming-pass promotion, observability, and an executable selftest harness.

/crons

Crons

The content engine and background jobs that run on a schedule, with payloads checked into the repo.

Quick start

Clone and bootstrap

# clone it git clone https://github.com/prasithg/clawrari.git cd clawrari # run the bootstrap ./bootstrap/init.sh # then read the build playbook open docs/playbook.md

The bootstrap generates the core files from templates. The playbook covers connectors, cron setup, night work, and upgrades.

Proof

Real numbers

8
production skills
44
docs pages
4
executable CI selftests
v0.5.0
current release

55 commits since February 2026, work landing on 15 of the last 30 days. Counts verified from the repo tree; see the changelog and live CI workflow ↗.