Skip to content

MAS Lab — Packages Reference

Auto-generated from pyproject.toml files and library.yaml manifests. Run task docs-gen to refresh after adding or modifying packages.

Summary

Package Layer Description CLI
mas-runtime Runtime core MAS Runtime V2 — Mealy kernel (embeddable library) mas-runtime
mas-ctl Orchestration MAS control plane V2 — compile, compose, session, placement mas-ctl
mas-lab Lab framework MAS Lab — Multi-Agent System experimentation, benchmarking, and analysis toolkit. mas-lab, mas-ctl
mas-lab-core Lab framework Core contracts, telemetry, schemas and utilities shared across MAS Lab components. —
mas-lab-bench Lab framework Benchmark engine, pipeline execution, plots and validation for MAS Lab. —
mas-lab-controller Lab framework MAS Lab controller daemon — workers, IPC, HTTP API for CLI and UI. —
mas-library-standard Libraries Standard library for MAS — infra bundles, tools, and pattern aliases. —
mas-library-skills Libraries Skills library for MAS — ContextContract catalog plugin + ToolContract skill-access tools. —
mas-library-eval Libraries Evaluation and quality metrics library for MAS Lab — MCE integration. —
mas-library-lab Libraries MAS Lab provider library with public eval plugins. —
mas-library-samples Libraries Community-contributed MAS sample apps, datasets, and tools. —
mas-library-ioa Libraries MCP client/server bridge for MAS Lab mas-mcp
mas-library-telemetry Libraries MAS Lab native-events → OTel conversion, span contracts/verification, and OTLP collector serialization. —
mas-library-kg Libraries MAS Lab trace normalization and knowledge-graph validation algorithms. —
---

Installation

Minimal install (core toolchain)

Installs the four core packages needed to run and orchestrate agents, execute benchmarks, and call mas-lab.

# From the repository root (uses UV_PROJECT_ENVIRONMENT if set):
task install
# Equivalent to:
uv pip install -e runtime -e ctl -e library-standard -e lab

Full install (all packages + optional extras)

This is not a minimal install — it covers every package in this repository plus all optional feature extras. Use this when you need evaluation metrics, or community sample apps.

uv pip install \
  -e runtime \
  -e ctl \
  -e library-standard \
  -e lab \
  -e library-eval \
  -e library-lab \
  -e library-samples

See Tutorial 0 — Environment Setup for the complete walkthrough including LLM endpoint wiring and verification.

Lab-specific backends

Third-party backends used by a single lab (e.g. letta in labs/extensions.lab) are not extras of any package here — some of them conflict with core workspace dependencies. They are declared in <lab>/requirements.txt and installed into a dedicated venv:

task install-lab LAB=labs/extensions.lab

Extended OTel/KG observability is not an OSS extra — see mas-lab-internal.


Package Details

mas-runtime

Install path: runtime
Layer: Runtime core
MAS Runtime V2 — Mealy kernel (embeddable library)

uv pip install -e runtime  # source
uv pip install "mas-runtime[grpc,dev]"  # PyPI, all extras

Optional extras:

Extra Packages / features
grpc grpcio>=1.60.0, grpcio-tools>=1.60.0
dev pytest>=8.0, pytest-asyncio>=0.24

CLI commands:

Command Entry point
mas-runtime mas.runtime.cli.main:main

mas-ctl

Install path: ctl
Layer: Orchestration
MAS control plane V2 — compile, compose, session, placement

uv pip install -e ctl  # source
uv pip install "mas-ctl[curses,dev]"  # PyPI, all extras

Depends on: mas-runtime, mas-library-standard, mas-library-skills

Optional extras:

Extra Packages / features
curses windows-curses>=2.3.2; sys_platform == 'win32'
dev pytest>=8.0, import-linter>=2.0

CLI commands:

Command Entry point
mas-ctl mas.ctl.cli.main:main

mas-lab

Install path: lab
Layer: Lab framework
MAS Lab — Multi-Agent System experimentation, benchmarking, and analysis toolkit.

uv pip install -e lab  # source
uv pip install mas-lab  # PyPI

Depends on: mas-runtime, mas-ctl, mas-lab-core, mas-lab-bench, mas-lab-controller, mas-library-eval

CLI commands:

Command Entry point
mas-lab mas.lab.cli:app
mas-ctl mas.ctl.cli.main:main

mas-lab-core

Install path: lab/components/core
Layer: Lab framework
Core contracts, telemetry, schemas and utilities shared across MAS Lab components.

uv pip install -e lab/components/core  # source
uv pip install mas-lab-core  # PyPI

Depends on: mas-runtime

Python entry point mas.lab.controller.plugins: this package advertises 1 discoverable item(s) so the runtime can load lab.controller.plugins without a hard-coded path.


mas-lab-bench

Install path: lab/components/bench
Layer: Lab framework
Benchmark engine, pipeline execution, plots and validation for MAS Lab.

uv pip install -e lab/components/bench  # source
uv pip install mas-lab-bench  # PyPI

Depends on: mas-lab-core, mas-library-eval, mas-runtime, mas-ctl

Python entry point mas.lab.runners: this package advertises 1 discoverable item(s) so the runtime can load lab.runners without a hard-coded path.


mas-lab-controller

Install path: lab/components/controller
Layer: Lab framework
MAS Lab controller daemon — workers, IPC, HTTP API for CLI and UI.

uv pip install -e lab/components/controller  # source
uv pip install "mas-lab-controller[dev]"  # PyPI, all extras

Depends on: mas-lab-core, mas-lab-bench, mas-runtime, mas-ctl

Optional extras:

Extra Packages / features
dev pytest>=9.0, pytest-asyncio, pytest-cov>=6.0, httpx>=0.27

Python entry point mas.lab.cli.components: this package advertises 1 discoverable item(s) so the runtime can load lab.cli.components without a hard-coded path.


mas-library-standard

Install path: library-standard
Layer: Libraries
Standard library for MAS — infra bundles, tools, and pattern aliases.

uv pip install -e library-standard  # source
uv pip install mas-library-standard  # PyPI

Depends on: mas-runtime

Python entry point mas.runtime.manifest_libraries: this package advertises 1 discoverable item(s) so the runtime can load runtime.manifest_libraries without a hard-coded path.


mas-library-skills

Install path: library-skills
Layer: Libraries
Skills library for MAS — ContextContract catalog plugin + ToolContract skill-access tools.

uv pip install -e library-skills  # source
uv pip install "mas-library-skills[all]"  # PyPI, all extras

Depends on: mas-runtime

Optional extras:

Extra Packages / features
adk google-adk>=1.0.0,<3.0.0
langchain deepagents>=0.7.0,<1.0.0
all google-adk>=1.0.0,<3.0.0, deepagents>=0.7.0,<1.0.0

Python entry point mas.runtime.manifest_libraries: this package advertises 1 discoverable item(s) so the runtime can load runtime.manifest_libraries without a hard-coded path.


mas-library-eval

Install path: library-eval
Layer: Libraries
Evaluation and quality metrics library for MAS Lab — MCE integration.

uv pip install -e library-eval  # source
uv pip install "mas-library-eval[dev]"  # PyPI, all extras

Depends on: mas-lab-core, mas-lab-bench, mas-runtime

Optional extras:

Extra Packages / features
dev pytest>=7.0.0, pytest-asyncio>=0.21.0

Python entry point mas.lab.cli.components: this package advertises 1 discoverable item(s) so the runtime can load lab.cli.components without a hard-coded path.

Python entry point mas.runtime.manifest_libraries: this package advertises 1 discoverable item(s) so the runtime can load runtime.manifest_libraries without a hard-coded path.


mas-library-lab

Install path: library-lab
Layer: Libraries
MAS Lab provider library with public eval plugins.

uv pip install -e library-lab  # source
uv pip install mas-library-lab  # PyPI

Depends on: mas-lab

Python entry point mas.lab.eval.plugins: this package advertises 1 discoverable item(s) so the runtime can load lab.eval.plugins without a hard-coded path.

Python entry point mas.runtime.manifest_libraries: this package advertises 1 discoverable item(s) so the runtime can load runtime.manifest_libraries without a hard-coded path.


mas-library-samples

Install path: library-samples
Layer: Libraries
Community-contributed MAS sample apps, datasets, and tools.

uv pip install -e library-samples  # source
uv pip install mas-library-samples  # PyPI

Python entry point mas.runtime.manifest_libraries: this package advertises 1 discoverable item(s) so the runtime can load runtime.manifest_libraries without a hard-coded path.


mas-library-ioa

Install path: library-ioa
Layer: Libraries
MCP client/server bridge for MAS Lab

uv pip install -e library-ioa  # source
uv pip install "mas-library-ioa[all]"  # PyPI, all extras

Depends on: mas-runtime

Optional extras:

Extra Packages / features
cli mcp[cli]
all mcp[cli]

CLI commands:

Command Entry point
mas-mcp library_ioa.plugins.mcp.server:main

Python entry point mas.runtime.manifest_libraries: this package advertises 1 discoverable item(s) so the runtime can load runtime.manifest_libraries without a hard-coded path.


mas-library-telemetry

Install path: library-telemetry
Layer: Libraries
MAS Lab native-events → OTel conversion, span contracts/verification, and OTLP collector serialization.

uv pip install -e library-telemetry  # source
uv pip install "mas-library-telemetry[all]"  # PyPI, all extras

Optional extras:

Extra Packages / features
convert opentelemetry-sdk>=1.20, opentelemetry-exporter-otlp-proto-http>=1.20, ioa-observe-sdk>=1.0.49
observe ioa-observe-sdk>=1.0.49
verify pyyaml>=6.0.0, jsonschema>=4.0.0
clickhouse clickhouse-connect>=0.7
bench mas-lab-bench, click>=8.0
dev pytest>=7.0, pytest-cov>=4.0, opentelemetry-sdk>=1.20, ioa-observe-sdk>=1.0.49, pyyaml>=6.0.0, jsonschema>=4.0.0, click>=8.0
all mas-library-telemetry[dev,bench,clickhouse]

Python entry point mas.runtime.manifest_libraries: this package advertises 1 discoverable item(s) so the runtime can load runtime.manifest_libraries without a hard-coded path.

Python entry point mas.lab.cli.components: this package advertises 1 discoverable item(s) so the runtime can load lab.cli.components without a hard-coded path.


mas-library-kg

Install path: library-kg
Layer: Libraries
MAS Lab trace normalization and knowledge-graph validation algorithms.

uv pip install -e library-kg  # source
uv pip install "mas-library-kg[all]"  # PyPI, all extras

Optional extras:

Extra Packages / features
ontology oxp-ontology[models]>=1.0.1
graph rdflib>=6.0.0, pyshacl>=0.25.0
validation oxp-ontology[models]>=1.0.1, rdflib>=6.0.0, pyshacl>=0.25.0
verify jsonschema>=4.0.0, pyyaml>=6.0.0
neo4j neo4j>=5.0
bench mas-lab-bench, click>=8.0
plot mas-lab-bench, mas-lab
norm norm @ git+https://github.com/outshift-open/observe-and-explain-platform.git@main#subdirectory=backend/norm
compare-native-otel mas-library-telemetry
dev pytest>=7.0, pytest-cov>=4.0, pytest-asyncio>=0.21, click>=8.0, ruff>=0.12.10, oxp-ontology[models]>=1.0.1, rdflib>=6.0.0, pyshacl>=0.25.0, jsonschema>=4.0.0, pyyaml>=6.0.0
all mas-library-kg[dev,neo4j,norm]

Python entry point mas.lab.cli.components: this package advertises 1 discoverable item(s) so the runtime can load lab.cli.components without a hard-coded path.

Python entry point mas.lab.processors: this package advertises 1 discoverable item(s) so the runtime can load lab.processors without a hard-coded path.

Python entry point mas.runtime.manifest_libraries: this package advertises 1 discoverable item(s) so the runtime can load runtime.manifest_libraries without a hard-coded path.