MPS Engine

The official engine implementation of Context Meta Pattern Strategy (MPS) architecture.

Context is an extension of the MPS architecture built on top of Markitdown, its whole purpose is to allow you to freely input any kind of file and the mps-engine automatically fit it into the MPS architecture.

MPS

from mps import Mps, MPSHierarchy

mps = MPS()
# uses default hierarchy, you can import it and see it at <mps.MPSLocatorDefaults>

# or
mps = Mps.from_location("path/to/project/mps")

# or
mps = Mps.from_location(MPSHierarchy(
    base_dir = "mps",
    meta_dir = "meta",
    pattern_dir = "pattern",
    context_dir = "context",
    strategy_dir = "strategy"
))

Context

Example

from mps import context

# => Markdown file placed under mps/context/my_finances.md
c = context(name="my-finances")  # Context object

Supported Formats

• xml
• pdf
• html
• .mp4
• .mp3

Pattern

Example

from mps import pattern, inject

ta = pattern(name="teacher-assistant")  # => Pattern object

# or if a variable exist in the pattern
p = inject(pattern("translate"), lang_code="en-US")

Strategy

Example

from mps import strategy

cot = strategy(name="CoT")  # => Strategy object

Integration

Semantic Kernel

1. Supplying kernel functions with more context

@context(name="my-details")
@kernel_function(name="greeter", description="Greets people")
def greet(name: str) -> str:
    return f"Hello {name}"

2. Personalizing agents

from semantic_kernel.agents import ChatCompletionAgent
from mps import pattern

p = pattern("student-tutor")
# or
p = pattern("www.mydrive.com/student-tutor.md")
# or
p = pattern("www.mydrive.com/student-tutor.md", nopersist=True)  #* (not recommended)
# by default the engine thinks the url that you used for a pattern is useful so it persists
# this is due to the [reusability philosophy](<../../README.md#reusability-philosophy>)
# and the [insurance phiolosophy](<../../README.md#insurance-philosophy>)
# in the MPS architecture

student_tutor_agent = ChatCompletionAgent(
    instructions=pattern("student-tutor"),
)

3. Giving agents a personality

from semantic_kernel.agents import ChatCompletionAgent
from mps import strategy, strategizer, context, pattern

finance_agent = ChatCompletionAgent(
    instructions=strategizer(
        context=context("my-monthly-billings"),
        pattern=pattern("finance-consultant"),
        strategy=startegy("AoT"),
    )
)

Interfaces (Miniatures & MPS)

Miniatures

from mps import get_minatures_homeland, set_minatures_homeland, Miniatures

get_minatures_homeland()
set_minatures_homeland("path/to/project/miniatures/")

# or

Miniatures.homeland = "path/to/project/miniatures/"
print(Miniatures.homeland)

MPS

from mps import get_mps_basedir, set_mps_basedir, Mps

get_mps_basedir()
set_mps_basedir("path/to/project/mps/")


Mps.basedir = "/path/to/project/mps"
print(Mps.basedir)

Roadmap

The following features are planned for future releases:

1. Complete core functionality - Finish any NotImplementedError or @future decorated functions
2. Documentation sprint - Focus on comprehensive documentation (also including CHANGELOG.md and breaking changes)
3. Test coverage - Aim for 80%+ test coverage on core functionality
4. Community preparation - Code of conduct, contribution guidelines
5. CI/CD setup - GitHub Actions for testing, linting, and deployment to PyPI

Nice to have

1. AsyncIO Support: Full async API for all MPS operations
2. Cache Support: Cache frequently requested miniatures via HTTP
3. Compilation Support: Compile the mps/ directory into a python script to reduce io overhead, and a potentially use for direct imports.
4. MPS Interface: A friendly UI to show/filter/add/modify miniatures in your local mps collection.
5. MPS Shop: A remote website where you can pull miniatures made by community to your local mps collection.
6. MCP Integration: Have your local mps collection converted as MCP prompts.