Whinchat (LLM assistant)

ChatBlock ¶

ChatBlock(item_id: str | None = None, collection_id: str | None = None, init_data: dict | None = None, unique_id: str | None = None)

Bases: DataBlock

This block uses API calls to external LLMs via Langchain to provide a conversational interface to a user's data.

Implemented models include:

the GPT series of models from OpenAI
Claude from Anthropic

Each needs the server to be configured with the corresponding API keys:

OPENAI_API_KEY,
ANTHROPIC_API_KEY.

A discussion of this block can be found in:

Jablonka et al, Digital Discovery, 2023,2, 1233-1250, DOI: 10.1039/d3dd00113j

Parameters:

Name	Type	Description	Default
`item_id`	`str \| None`	The item to which the block is attached, or	`None`
`collection_id`	`str \| None`	The collection to which the block is attached.	`None`
`init_data`	`dict \| None`	A dictionary of data to initialise the block with.	`None`
`unique_id`	`str \| None`	A unique id for the block, used in the DOM and database.	`None`

Attributes:

Name	Type	Description
`accepted_file_extensions`
`block_db_model`
`block_id`
`blocktype`
`data`
`defaults`	`dict`
`description`
`multi_file`	`bool`	Whether this block can accept multiple files as input.
`name`
`plot_functions`
`version`	`str`	The implementation version of this particular block.

__supports_collections ¶

__supports_collections = True

accepted_file_extensions ¶

accepted_file_extensions = None

block_db_model ¶

block_db_model = ChatBlockResponse

block_id ¶

block_id = unique_id or generate_random_id()

blocktype ¶

blocktype = 'chat'

data ¶

data = {'item_id': item_id, 'collection_id': collection_id, 'blocktype': blocktype, 'block_id': block_id, None: defaults}

defaults ¶

defaults: dict = {'system_prompt': 'You are whinchat (lowercase w), a virtual data managment assistant that helps materials chemists manage their experimental data and plan experiments. You are deployed in the group of Professor Clare Grey in the Department of Chemistry at the University of Cambridge.\nYou are embedded within the program datalab, where you have access to JSON describing an ‘item’, or a collection of items, with connections to other items. These items may include experimental samples, starting materials, and devices (e.g. battery cells made out of experimental samples and starting materials).\nAnswer questions in markdown. Specify the language for all markdown code blocks. You can make diagrams by writing a mermaid code block or an svg code block. When writing mermaid code, you must use quotations around each of the labels (e.g. A["label1"] --> B["label2"])\nBe as concise as possible. When saying your name, type a bird emoji right after whinchat 🐦.\n        ', 'temperature': 0.2, 'model': 'gpt-4o', 'available_models': AVAILABLE_MODELS}

description ¶

description = 'Virtual LLM assistant block allows you to converse with your data.'

multi_file ¶

multi_file: bool = False

Whether this block can accept multiple files as input.

name ¶

name = 'Whinchat assistant'

plot_functions ¶

plot_functions

version ¶

version: str = __version__

The implementation version of this particular block.

continue_conversation ¶

continue_conversation(prompt: str | None) -> None

Continues the conversation based on the passed user prompt.

Parameters:

Name	Type	Description	Default
`prompt`	`str \| None`	The textual response from the user.	required

event_names ¶

event_names() -> set[str]

Return a list of event names supported by this block.

events_by_name ¶

events_by_name() -> dict[str, Callable]

Returns a dict of registered events for this block.

from_web ¶

from_web(data: dict)

Initialise the block state from data passed via web request with a given item, collection and block ID.

Parameters:

Name	Type	Description	Default
`data`	`dict`	The block data to initialiaze the block with.	required

null_event ¶

null_event(**kwargs)

A null debug event that does nothing but logs its kwargs and overwrites the data dict with the args.

process_events ¶

process_events(events: list[dict] | dict)

Handle any supported events passed to the block.

render ¶

render()

start_conversation ¶

start_conversation(item_data: dict | None = None, collection_data: dict | None = None)

Starts a new conversation with the system prompt, embedding the current item or collection data.

to_db ¶

to_db() -> dict

returns a dictionary with the data for this block, ready to be input into mongodb

to_web ¶

to_web() -> dict[str, Any]

Returns a JSON serializable dictionary to render the data block on the web.

update_from_web ¶

update_from_web(data: dict)

Update the block with validated data received from a web request. Will strip any fields that are "computed" or otherwise not controllable by the user.

Parameters:

Name	Type	Description	Default
`data`	`dict`	A dictionary of data to update the block with.	required