zms.llmtools

module documentation

(source)

llmtools.py – ZMS tool registry for LLM function/tool calling.

Provides:

``LLM_TOOLS``: built-in tool schemas (OpenAI-format, also accepted by Ollama v0.3+).
``execute_llmtool(name, args, context)``: built-in dispatcher for core ZMS tools.
``ZMSLLMToolsAdapter``: abstraction layer that can load custom toolsets from ``*_llmtools`` ZMSLibrary meta-objects and dispatch custom ``llmtool_*`` actions.

All tools operate on the metamodel manager reachable via the ZMS acquisition context.

License: GNU General Public License v2 or later, Organization: ZMS Publishing

Class	`ZMSLLMToolsAdapter`	Resolve active LLM tools connector and dispatch tool calls.
Function	`execute_llmtool`	Dispatch a tool call to the corresponding ZMS API.
Function	`get_available_llmtools_connectors`	Return available custom LLM tools connectors (ZMSLibrary meta-objects).
Constant	`LLM_TOOLS`	Undocumented
Function	`_tool_index_qdrant`	Crawl every ZMS page, render its body content, chunk + embed the text, and upsert all vectors into the configured Qdrant collection.

def execute_llmtool(name, args, context): (source) ¶

Dispatch a tool call to the corresponding ZMS API.

Parameters
name	Tool name as returned in the LLM tool_call.
args	Dict of parsed arguments from the LLM.
context	ZMS acquisition context (must have ``metaobj_manager`` attribute).
Returns
JSON-serialisable result dict.

def get_available_llmtools_connectors(context): (source) ¶

Return available custom LLM tools connectors (ZMSLibrary meta-objects).

A connector is any ``ZMSLibrary`` matching either:

id ends with ``_llmtools`` (generic naming), or
id ends with ``_connector`` and package starts with ``com.zms.llmtools.`` (connector-style naming, e.g. ``ollama_connector``).

LLM_TOOLS: list[dict] = (source) ¶

Undocumented

Value

[{'type': 'function',
  'function': {'name': 'list_content_types',
               'description': 'Return the list of all content type IDs defined i↵
n the ZMS metamodel. Use this first to discover what types already exist before ↵
creating new ones.',
               'parameters': {'type': 'object',
                              'properties': {},
...

def _tool_index_qdrant(args, context): (source) ¶

Crawl every ZMS page, render its body content, chunk + embed the text, and upsert all vectors into the configured Qdrant collection.

Runs fully inside the Zope process — no external HTTP calls required.