llamaclaw — the umbrella¶

Lean, polyglot, terminal-first scientific computing — one clone gets you everything.¶

Sub-repos · Quick start · Workflow · Scripts · License

Umbrella repo for the llamaclaw ecosystem — a lean, polyglot, terminal-first scientific computing stack for epidemiological and statistical modeling.

All sub-repos are private during the incubation phase. They are pulled in here as git submodules so a single clone gives you the whole workspace.

Sub-repos¶

Repo	What	Language
`esml/`	Core Python + R statistical package (causal inference, DML, epi)	Python + R + C
`turboquant/`	KV-cache compression (PolarQuant + QJL + TurboQuant MSE)	Python + C
`perseus/`	LLM relay — Ollama + FreeAPI + Gemini + OpenAI chain	Python
`autoresearch/`	Autonomous LLM pretraining (MPS main; CUDA `upstream`)	Python
`tide/`	Terminal IDE for ESML — x86-first, Alpine Docker	Go
`kronos/`	ARM64 terminal multiplexer for Raspberry Pi	Rust
`luci/`	Pi-side AI agent (OpenClaw/ZeroClaw orchestration)	Shell + Markdown
`yoda/`	Claude co-architect persona + agent specs + journal	Markdown
`deploy/`	OS deploy configs (mac launchd + pi systemd/kernel/udev)	Shell + unit files
`installer/`	pip + Homebrew + curl\|sh; Homebrew tap	Shell + Ruby
`intersphinx/`	Ecosystem docs — landing, tutorials, deploy guides	Sphinx + Markdown
`parallelism/`	Coord protocol for parallel Yodas — scripts, schedulers, install.sh	Shell
`propagator/`	Cross-repo infrastructure distribution (boot hooks, parity audits)	Shell

Plus the Ollama org for Perseus model weights.

Quick start¶

git clone --recurse-submodules git@github.com:llamaclaw/llamaclaw.git
cd llamaclaw
./scripts/init.sh          # pulls LFS, sets up venvs, verifies toolchain

Or, after cloning without submodules:

git submodule update --init --recursive

Workflow¶

Every sub-repo follows dev → main via PR:

dev is the default branch — every push lands there first.
CI runs on every push to dev and every PR to main.
When ready, open a PR dev → main; merge it; main advances.

Work inside a submodule:

cd esml
git checkout dev
# ...edits...
git push                   # triggers CI on llamaclaw/esml

When a submodule’s dev tip has been pinned (via PR to main), update the umbrella:

# from the umbrella root
git submodule update --remote esml
git add esml
git commit -m "bump: esml → <sha>"
git push

Scripts¶

Script	Purpose
`scripts/init.sh`	Clone + init submodules; install venvs; verify toolchain
`scripts/release.sh`	Coordinated version-bump across submodules (tagged releases)
`scripts/update-all.sh`	Pull latest `dev` on every submodule
`scripts/status.sh`	Show state of each submodule (branch, ahead/behind, dirty)

License¶

Each sub-repo is GPL-3.0-or-later. This umbrella repo (orchestration scripts + docs) is also GPL-3.0-or-later. See LICENSE.

Parallel Yodas — coord system¶

When multiple Claude sessions (“Yodas”) run across this ecosystem, they coordinate via llamaclaw/parallelism. Scripts: coord-claim, coord-release, coord-check, coord-watch (SSH-tailed poke log on zeus). Boot step 8 of every session runs coord-check to see what’s claimed before starting work.