cmeta 0.29.1

cMeta (aka cX) is a common meta-framework for unifying, interconnecting and reusing code, data, models, agents, and knowledge across projects and domains.

cMeta (Common Meta Framework)

cMeta (also known as cX) is a small, portable framework for unifying, interconnecting, and reusing code, data, models, agents, and knowledge across projects, platforms, and time through a uniform interface. It enables collaborative research and experimentation for developing self-optimizing and self-adapting software and hardware that automatically identify the most efficient and cost-effective ways to execute AI, ML, and other complex workloads.

Created, architected and developed by Grigori Fursin — originator of the long-term vision, concept, architecture and successive prototypes behind cMeta.

Core features

cMeta represents each part of a workflow as a uniform, composable, content-addressed, reproducible artifact, reachable through a single interface:

• One uniform interface — cm.access({'category', 'command'}) in Python and cx <category> <command> on the command line reach everything: run a program, fetch a model, prepare a dataset, build a toolchain, invoke an agent.
• Composable automations — workflows are assembled from small, reusable tasks that declare what they use, instead of hard-coded scripts.
• Extensible & pluggable — new capabilities are added as self-contained artifacts (categories, tasks, tools) with optional Python hooks, so the framework grows by plugging in components rather than modifying the core.
• Metadata & tags — structured, machine-readable identity makes any component discoverable and reusable, by tags rather than by hard-coded paths.
• Content-addressed caching & better reproducibility — identical work is not repeated, and the full context of a run is captured to help reproduce it. Full determinism across heterogeneous environments is hard; cMeta improves reproducibility but does not yet fully solve it — this remains ongoing community R&D.
• Virtualized portability — toolchains, compilers, drivers and runtimes are detected, isolated and pinned, abstracting away OS and accelerator differences.
• Unifying agents — AI agents operate the same discovery, composition and execution surface that humans do, so automations can be driven by people and by agents through one interface.

---

Installation

pip install cmeta
cmeta --version
cx --version

See docs/installation.md for uv, install-from-source, configuration and troubleshooting.

---

Quickstart

Command line:

cx --help
cx repo list

Python:

from cmeta import CMeta

cm = CMeta()
r = cm.access({'category': 'repo', 'command': 'list'})
print(r)

---

License

Apache 2.0 — see LICENSE.

This project may include minor functionality reused from MLCommons CK/CM, developed by the same author and licensed under the same Apache 2.0 terms.

Copyright

Copyright (C) 2025–2026 Grigori Fursin and cTuning Labs.

Links

• Author: https://cTuning.ai/@gfursin
• Organizations: cTuning Labs and the cTuning foundation
• Artifact Evaluation and Reproducibility Initiatives

Status

Under active development.