Streamlit Web UI

Relevant source files

The following files were used as context for generating this wiki page:

• app.py
• README.md
• orchestrator.py
• scenarios.py
• memory_utils.py

The Streamlit Web UI serves as the primary operations dashboard for the BrightPath AI system. It provides a visual interface for managing the multi-agent pipeline, monitoring tasks, inspecting memory files, and viewing CRM data. The interface is designed for use by managers to oversee AI-driven lead handling and customer service automation.

The UI is structured around a four-tab layout that separates active pipeline execution from historical data and system state. It integrates directly with the Anthropic API for agent logic, a SQLite database for event logging, and local Markdown files for long-term agent memory.

Sources: README.md:1-25, app.py:11-20

Interface Architecture and Styling

The UI is built with a highly customized design language using the DM Sans, Syne, and DM Mono fonts. It implements a sophisticated CSS layer to provide distinct visual feedback for different agent states and message priorities.

Visual Components

The system uses a card-based layout (bp-card) with color-coded accents to denote specific statuses:

• Default Card: Standard white background with a grey border.
• Accent Card: Yellow border for Manager Briefing outputs.
• Green Card: Green border indicating successful Specialist Agent execution.
• Red/Blue Cards: Used for alerts and specific data highlighting.
• Tags: Micro-components for displaying category, urgency, and sentiment (e.g., tag-sales, tag-high, tag-positive).

Sources: app.py:41-160

Sidebar and Statistics

The sidebar acts as a persistent control and monitoring panel across all tabs. It includes:

1. Identity: BrightPath branding.
2. API Configuration: A password-masked input for the Anthropic API key.
3. Real-time Metrics: Four dynamic cards showing total Events, Open Tasks, High Urgency items, and CRM Updates.
4. Model Configuration: Display of the currently active model (e.g., claude-sonnet-4-5).

Sources: app.py:461-510

Main Navigation Tabs

The dashboard is divided into four primary functional areas defined in the main() function of app.py.

Tab	Purpose	Key Functionality
Run Pipeline	Execution	Select scenarios, run the 3-agent pipeline, and view real-time logs.
Tasks & History	Monitoring	View open tasks from the SQLite database and recent event history.
Memory Files	Inspection	Live view of the five .md files used by agents for context.
CRM Data	Data Management	Display contact information, VIP status, and estimated values.

Sources: README.md:118-125, app.py:512-515

Agent Pipeline Execution

The "Run Pipeline" tab facilitates the core workflow. Users can either select from 18 pre-built test scenarios or enter a custom message.

Execution Flow

The run_pipeline function manages the sequential execution of agents and the subsequent display of results in two main columns (col_left for input, col_right for status and results).

The diagram shows the logic flow from user input through the three-agent sequence (Triage, Specialist, Briefing) and the final data persistence layer.
Sources: app.py:342-450, orchestrator.py:270-350

Pipeline Steps

1. Step 1: Triage: Classifies the message and determines the next_agent. It renders tags for category and sentiment.
2. Step 2: Specialist: Executes either the Sales or Customer Service agent. If the category is admin, it builds a rich context from the database and memory for a manager briefing.
3. Step 3: Briefing: Generates a Telegram-ready summary and flags for value, risk, or human action.

Sources: app.py:348-408

Data Persistence and Memory Integration

The Streamlit UI performs heavy read/write operations to keep the UI in sync with the agent state.

SQLite Interaction

The dashboard pulls data from three primary tables: events, tasks, and crm_log. Functions like get_counts(), get_tasks(), and get_events() query the database to populate metric cards and history lists.

This diagram represents the schema of the brightpath.db file used by the Streamlit application to log all activity.
Sources: app.py:204-268, orchestrator.py:45-95

Memory Synchronization

After a pipeline run, the UI calls run_memory_update from memory_utils.py. This ensures that any decisions made (e.g., escalating a complaint or identifying a VIP upsell) are written back to the local Markdown files:

• customer_history.md
• sales_notes.md
• open_tasks.md
• decision_log.md
• company_context.md

Sources: memory_utils.py:1-40, app.py:415-417

Scenario Loading Logic

The "Run Pipeline" tab includes a selectbox populated from scenarios.py. When a scenario is selected, its corresponding message text is automatically loaded into the input area. This is handled by mapping the SCENARIOS dictionary keys to the UI component.

Sources: scenarios.py:10-100, app.py:525-535

The Streamlit Web UI acts as the central hub for the BrightPath system, providing a visual layer that abstracts the complex multi-agent orchestration into a manageable business dashboard. By integrating real-time agent processing with live database and file system updates, it provides a transparent view into the AI's decision-making process.