mdi-llmkit 1.1.2

Utilities for managing multi-shot conversations and structured data handling in LLM applications

mdi-llmkit

Utilities for managing LLM chat conversations and structured JSON responses with OpenAI's Responses API.

Installation

pip install mdi-llmkit

Quick Start

Preferred subpackage imports:

from mdi_llmkit.gpt_api import GptConversation
from mdi_llmkit.json_surgery import json_surgery

gpt_submit

from openai import OpenAI
from mdi_llmkit.gpt_api import gpt_submit

client = OpenAI()

reply = gpt_submit(
    messages=[{"role": "user", "content": "Say hello."}],
    openai_client=client,
)
print(reply)

GptConversation

from openai import OpenAI
from mdi_llmkit.gpt_api import GptConversation

client = OpenAI()
conversation = GptConversation(openai_client=client)

reply = conversation.submit_user_message("Give me three project name ideas.")
print(reply)

JSON Response Mode

from openai import OpenAI
from mdi_llmkit.gpt_api import gpt_submit

client = OpenAI()

result = gpt_submit(
    messages=[{"role": "user", "content": "Return JSON with keys a and b."}],
    openai_client=client,
    json_response=True,
)

print(type(result))  # dict or list
print(result)

JSON Surgery

from openai import OpenAI
from mdi_llmkit.json_surgery import json_surgery

client = OpenAI()

obj = {"status": "pending", "tags": ["alpha"]}
result = json_surgery(
    client,
    obj,
    'Set status to "approved" and append "done" to tags.',
)

print(result)

Placemark helpers are available in mdi_llmkit.json_surgery.placemarked_json:

• placemarked_json_stringify(obj, indent=2, skipped_keys=None)
• navigate_to_json_path(obj, json_path)

Semantic List Comparison (compare_item_lists)

Use compare_item_lists when you need to compare a before list and an after list and classify items as removed, added, renamed, or unchanged. This uses AI to resolve semantic renames that would otherwise appear as unrelated remove/add pairs in strict string comparison.

from openai import OpenAI
from mdi_llmkit.semantic_match import compare_item_lists

client = OpenAI()

result = compare_item_lists(
    client,
    ["Legacy Plan", "Shared Item"],
    ["Modern Plan", "shared item"],
    (
        "Legacy Plan was renamed to Modern Plan. "
        "Shared Item is unchanged."
    ),
)

print(result)
# [
#   {
#     "item": "Legacy Plan",
#     "classification": "renamed",
#     "new_name": "Modern Plan",
#   },
#   {
#     "item": "Shared Item",
#     "classification": "unchanged",
#     "new_name": None,
#   },
# ]

Accepted input item formats:

• String item name: "Item Name"
• Object with optional context: {"name": "Item Name", "description": "..."}

Notes:

• Name matching is case-insensitive for exact unchanged detection.
• Names should be unique within each input list (case-insensitive).
• Optional description is context-only and does not define identity.

Local Dev (Windows venv)

From packages/python-mdi-llmkit, activate the project venv and run tests:

.\venv\Scripts\Activate.ps1
python -c "import sys; print(sys.executable)"
python -m unittest tests/test_placemarked_json.py tests/test_json_surgery_unit.py

Live json_surgery integration tests (real API) require OPENAI_API_KEY:

python -m unittest tests/test_json_surgery.py

Notes

• Package name for pip install is mdi-llmkit.
• Python import package is mdi_llmkit.
• gpt_submit supports optional warning reporting via warning_callback.