Implementation:Truera Trulens UX Components

Knowledge Sources	Truera_Trulens TruLens Docs
Domains	Dashboard, Visualization
Last Updated	2026-02-14 08:00 GMT

Overview

Streamlit UI components module providing reusable functions for rendering selectors, call frames, metadata, LLM details, agent details, tool details, and prompt information in the TruLens dashboard.

Description

The components module in trulens.dashboard.ux provides a collection of Streamlit rendering functions used throughout the TruLens dashboard to display application internals, call traces, and component metadata. Key functions include:

Content Display:

write_or_json(st, obj) -- Smart dispatcher that renders a string as st.json if it parses into a non-string JSON structure (e.g., a dict), or as st.write otherwise. Handles parse failures gracefully by falling back to st.write.

Selector Rendering:

copy_to_clipboard(path) -- Stores a selector path string in st.session_state.clipboard for copy-to-clipboard functionality.
draw_selector_button(path) -- Renders a Streamlit button labeled with the dashboard-formatted selector path. Clicking the button copies the path to the clipboard. Uses random.random() for unique widget keys.
render_selector_markdown(path) -- Returns a markdown-formatted string with the selector path rendered for dashboard display, enclosed in backticks and brackets.

Call Frame Rendering:

render_call_frame(frame, path) -- Formats a RecordAppCall frame as a markdown string showing the method name and its fully-qualified class path (module and class name), using bold formatting.

Metadata Display:

dict_to_md(dictionary) -- Converts a dictionary into a Markdown table string with a single header row and single data row. Returns "No metadata." for empty dictionaries.
draw_metadata_and_tags(metadata, tags) -- Merges tags into the metadata dictionary, renames internal column names (e.g., EXTERNAL_APP_COL_NAME to "External App?", PINNED_COL_NAME to "Pinned?") for user-friendliness, and renders as a Markdown table.

Call Tree Rendering:

draw_call(call) -- Renders a single RecordAppCall as a Streamlit expander. The expander label shows the method name, class, and selector path. Inside, it displays the call stack (via frames), input arguments (as JSON or write), and output returns.
draw_calls(record, index) -- Iterates through all calls in a record and renders only the call matching the given step index using draw_call.

Component Detail Panels:

draw_prompt_info(query, component) -- Renders prompt details for a component as expandable sections, filtering out None, empty, and non-serializable values. Each section is labeled with its key and the corresponding selector path.
draw_llm_info(query, component) -- Renders LLM component details as a Streamlit table. Handles nested dictionary columns by flattening them into separate columns. Redacts columns whose names indicate secrets (via key_utils.should_redact_key).
draw_agent_info(query, component) -- Renders agent details using the same pattern as draw_prompt_info.
draw_tool_info(query, component) -- Renders tool details using the same pattern as draw_prompt_info.

Usage

Use these functions when building custom dashboard pages that need to display TruLens application internals. The draw_call and draw_calls functions are central to the record detail/trace views. The draw_llm_info, draw_prompt_info, draw_agent_info, and draw_tool_info functions are used by the app detail pages to render component configurations.

Code Reference

Source Location

Repository: Truera_Trulens
File: src/dashboard/trulens/dashboard/ux/components.py
Lines: 1-327

Signature

def write_or_json(st, obj): ...


def copy_to_clipboard(path, *args, **kwargs): ...


def draw_selector_button(path) -> None: ...


def render_selector_markdown(path) -> str: ...


def render_call_frame(
    frame: record_schema.RecordAppCall, path=None,
) -> str: ...


def dict_to_md(dictionary: dict) -> str: ...


def draw_metadata_and_tags(
    metadata: types_schema.Metadata, tags: types_schema.Tags,
) -> str: ...


def draw_call(call: record_schema.RecordAppCall) -> None: ...


def draw_calls(record: record_schema.Record, index: int) -> None: ...


def draw_prompt_info(
    query: serial_utils.Lens, component: core_app.ComponentView,
) -> None: ...


def draw_llm_info(
    query: serial_utils.Lens, component: core_app.ComponentView,
) -> None: ...


def draw_agent_info(
    query: serial_utils.Lens, component: core_app.ComponentView,
) -> None: ...


def draw_tool_info(
    query: serial_utils.Lens, component: core_app.ComponentView,
) -> None: ...

Import

from trulens.dashboard.ux.components import write_or_json
from trulens.dashboard.ux.components import draw_selector_button
from trulens.dashboard.ux.components import render_selector_markdown
from trulens.dashboard.ux.components import render_call_frame
from trulens.dashboard.ux.components import draw_metadata_and_tags
from trulens.dashboard.ux.components import draw_call
from trulens.dashboard.ux.components import draw_calls
from trulens.dashboard.ux.components import draw_prompt_info
from trulens.dashboard.ux.components import draw_llm_info
from trulens.dashboard.ux.components import draw_agent_info
from trulens.dashboard.ux.components import draw_tool_info

I/O Contract

Inputs

write_or_json:

Name	Type	Required	Description
st	module	yes	The Streamlit module reference (passed explicitly).
obj	Any	yes	The object to display. If a JSON-parseable string that yields a dict, rendered as `st.json`.

draw_selector_button:

Name	Type	Required	Description
path	serial_utils.Lens	yes	A Lens selector path to display on the button and copy to clipboard.

render_call_frame:

Name	Type	Required	Description
frame	record_schema.RecordAppCall	yes	A single call frame from the record call stack.
path	Optional[serial_utils.Lens]	no	Override path. Defaults to `frame.path` if not provided.

draw_call:

Name	Type	Required	Description
call	record_schema.RecordAppCall	yes	The call record to render with its stack, args, and rets.

draw_calls:

Name	Type	Required	Description
record	record_schema.Record	yes	The full record containing all calls.
index	int	yes	The 1-based step index of the call to render.

draw_llm_info / draw_prompt_info / draw_agent_info / draw_tool_info:

Name	Type	Required	Description
query	serial_utils.Lens	yes	The Lens path identifying this component within the app.
component	core_app.ComponentView	yes	The component view containing the JSON details to display.

draw_metadata_and_tags:

Name	Type	Required	Description
metadata	types_schema.Metadata	yes	The metadata dictionary for the app.
tags	types_schema.Tags	yes	The tags string for the app.

Outputs

Name	Type	Description
write_or_json	None	Renders content directly to the Streamlit page.
draw_selector_button	None	Renders a button to the Streamlit page; copies path to session state on click.
render_selector_markdown	str	Returns a markdown-formatted selector string (e.g., [`Select.RecordCalls.retrieve`]).
render_call_frame	str	Returns a markdown string describing the call frame method and class.
dict_to_md	str	Returns a Markdown table string representation of a dictionary.
draw_metadata_and_tags	str	Returns a Markdown table string with metadata and tags combined.
draw_call / draw_calls	None	Renders call details directly to the Streamlit page.
draw_llm_info / draw_prompt_info / draw_agent_info / draw_tool_info	None	Renders component detail panels directly to the Streamlit page.

Usage Examples

from trulens.dashboard.ux.components import write_or_json
import streamlit as st

# Display a JSON string as formatted JSON
write_or_json(st, '{"key": "value", "nested": {"a": 1}}')

# Display a plain string as text
write_or_json(st, "This is a plain text response.")

from trulens.dashboard.ux.components import draw_calls

# Render a specific call step from a record
record = session.get_record(record_id="rec_abc123")
draw_calls(record, index=2)  # Renders the 2nd call step

from trulens.dashboard.ux.components import draw_metadata_and_tags

# Render metadata as a help tooltip
metadata = {"model": "gpt-4", "temperature": 0.7}
tags = "production,v2"
help_text = draw_metadata_and_tags(metadata, tags)
st.header("My App v2", help=help_text)

from trulens.dashboard.ux.components import draw_llm_info
from trulens.core.schema.select import Select

# Render LLM details for a specific component
query = Select.App.llm
component = app_json["components"][0]
draw_llm_info(query, component)

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment