BACR Website Documentation

Comprehensive documentation for website content based on the BACR application

Getting Started
Installation
CLI Reference
API Reference

// Getting Started

Overview

BACR (Bot Assistant with Contextual Retrieval) is a sophisticated AI chat assistant that combines advanced document processing, intelligent query routing, and persistent memory. Built with Python, Flet, LangGraph, and FAISS, BACR provides a powerful desktop AI assistant experience with local document understanding capabilities.

Key Highlights:

🤖 Multi-model AI routing with automatic query classification
📚 Hybrid RAG system with vector search and AI reranking
💾 Cross-session memory that learns from your conversations
🎨 Modern desktop GUI with theme support
🔐 Secure Google OAuth authentication
📊 Real-time token usage and cost tracking

Why BACR?

For Developers:

Local document processing without sending files to external APIs
Intelligent code-aware routing for programming questions
Multi-format support (PDF, DOCX, TXT, MD, and more)
Persistent conversation history across sessions

For Researchers:

Advanced RAG with Jina AI reranking for accurate document retrieval
Citation tracking - only cited sources are shown
Cross-session memory for building knowledge over time
Export conversations to PDF for documentation

For Teams:

User-isolated databases for privacy
Server-managed API keys for easy deployment
Zero Data Retention (ZDR) mode for privacy-conscious users
Cost tracking per query for budget management

Features

AI & Intelligence

Two-Stage Smart Routing: Queries are automatically classified and routed to the optimal model
- Stage 1: File management → RAG (document search) → Direct LLM
- Stage 2: Coding → Reasoning → Chat → Research
Models Integration: Access to 200+ AI models including GPT-4, Claude, Gemini, DeepSeek, and more
Real-time Streaming: Token-by-token responses with live status updates
Cost Tracking: Monitor token usage and costs per query with detailed breakdowns
Model Modes: Switch between ZDR and Non-ZDR

Document Processing (RAG)

Hybrid Search: Combines vector similarity with Jina AI reranking
Local Embeddings: Uses a local embedding GGUF model - no API calls for embeddings
Smart Filtering: Relevance threshold automatically removes irrelevant chunks
Intelligent Citations: LLM cites sources as (Doc 1), (Doc 2); only cited sources are displayed
Persistent Storage: User-specific FAISS indexes saved to ~/.bacr/user_databases/
Multi-Format Support: PDF, TXT, MD, DOCX, XLSX, PPTX, RTF, CSV, JSON, YAML, and more

Memory & Context

Cross-Session Memory: LangMem semantic memory that persists across all conversations
User-Specific Databases: Isolated SQLite databases per user for privacy
Persistent History: All conversations saved and searchable
Context-Aware: Each session maintains independent context with file tracking

User Interface

Modern Flet GUI: Clean, responsive desktop interface built with Flutter
Theme Support: Light, dark, and system themes (persists across restarts)
Tabbed Sessions: Multiple independent conversations with easy switching
Font Size Control: Adjustable text size with visual controls
Status Bar: Real-time file count, token usage, and credit balance
Export Mode: Select and export specific messages to PDF

Authentication & Security

Google OAuth: Secure login with auto-login support using refresh tokens
Profile Pictures: Fetched fresh from Google on each login
User Isolation: All data separated by user account
Privacy Protection: User data is isolated

Architecture

BACR is built with a modular architecture that separates concerns and enables easy maintenance:

Frontend Layer:

Flet-based GUI with reactive components
Google OAuth integration for authentication
Real-time status updates and streaming responses
Theme and preference management

Core Layer:

LangGraph-based conversation management
Two-stage intelligent routing system
Hybrid RAG with vector search and reranking
Cross-session memory with LangMem

Backend Layer:

Server API client for authentication and API key management
Model manager with ZDR/Non-ZDR mode switching
Multi-model access
Token usage tracking and cost calculation

Storage Layer:

User-specific SQLite databases
FAISS vector indexes for document embeddings
Local file system for document caching

Installation

Requirements

System Requirements:

Operating System: Windows 10/11, Linux (Ubuntu 20.04+, Fedora, Arch), macOS 11+
Python: 3.9 or higher
RAM: 4GB minimum, 8GB recommended
Disk Space: 2GB for application and dependencies
Internet: Required for authentication and API access

Windows

Linux

MacOS

Not available

CLI Reference

Commands

BACR provides a comprehensive set of slash commands for managing files, sessions, and conversations.

File Management

/add <filepath>

Add a document to the RAG system for context
Supports: PDF, TXT, MD, DOCX, XLSX, PPTX, RTF, CSV, JSON, YAML
Example: /add C:\Documents\report.pdf
Creates embeddings and indexes document for search

/remove <filename>

Remove a specific file from the RAG system
Example: /remove report.pdf
Removes embeddings and frees memory

/files

List all files currently in the RAG system
Shows file names and chunk counts
Displays total context size

/clear

Remove all files from the RAG system
Clears all document embeddings
Does not clear chat history

Session Management

/new-chat or Ctrl+N

Start a new conversation in a new tab
Creates independent context
Previous sessions remain accessible

/close-tab or Ctrl+W

Close the current conversation tab
Saves conversation to database
Cannot close last tab

/list-chats

Show all saved conversations
Displays session names and timestamps
Shows message counts

/load-chat <name_or_id>

Load a conversation into the current tab
Replaces current conversation
Example: /load-chat Project Discussion

/open-chat <name_or_id>

Open a conversation in a new tab
Keeps current conversation
Example: /open-chat Meeting Notes

/rename-chat <new_name>

Rename the current conversation
Example: /rename-chat Q1 Planning
Cannot rename temporary sessions

/delete-chat <name_or_id>

Delete a saved conversation permanently
Example: /delete-chat Old Notes
Cannot be undone

Chat History

/clear-chat or /clear

Clear the chat history for current conversation
Removes all messages from display and state
Does not delete saved conversation
Files remain in context

/export-chat

Export current conversation to a file
Saves as timestamped text file
Includes all messages and metadata
Location: ~/.bacr/exports/

/import-chat <filepath>

Import a conversation from a file
Supports TXT and PDF formats
Adds to current conversation
Example: /import-chat notes.txt

Utilities

/tokens

Show token usage for current conversation
Displays input, output, and total tokens
Shows estimated costs
Aggregates across all sessions

/find <query>

Search across all conversation history
Opens search interface
Highlights matching messages
Example: /find API documentation

/help

Opens help page in browser
Shows all available commands
Includes examples and descriptions

/quit

Close the application
Saves all conversations
Clears API keys from memory

Configuration

Theme Toggle

Click theme icon in top bar
Switches between light/dark
Preference persists across restarts

Font Size

Click font size buttons in top bar
Adjusts text size for all messages
Range: 12px to 20px

ZDR Mode Toggle

Click ZDR toggle in settings
Switches between free and premium models
Clears model cache on switch

Shell Commands

BACR supports common shell commands for file system navigation:

cd <directory>

Change current working directory
Example: cd Documents
Affects file operations

ls or dir

List files in current directory
Shows file names and sizes

pwd

Print working directory
Shows current path

cat <file>

Display file contents
Example: cat notes.txt

mkdir <directory>

Create new directory
Example: mkdir projects

rm <file>

Remove file (use with caution)
Example: rm temp.txt

cp <source> <dest>

Copy file
Example: cp file.txt backup.txt

mv <source> <dest>

Move or rename file
Example: mv old.txt new.txt

etc

Usage Examples

Basic Chat:

User: What is machine learning?
BACR: [Streams response with real-time status]

Document Processing:

User: /add C:\Reports\Q4_Analysis.pdf
BACR: ✔︎ Successfully added Q4_Analysis.pdf (45 chunks)

User: What were the key findings in Q4?
BACR: According to the Q4 Analysis (Doc 1), the key findings were...

Sources:
• Q4_Analysis.pdf

Multi-Session Workflow:

User: /new-chat
User: /add project_spec.md
User: Explain the authentication flow
BACR: [Provides detailed explanation with citations]

User: /rename-chat Auth Documentation
User: /export-chat
BACR: ✔︎ Exported to ~/.bacr/exports/Auth_Documentation_2026-03-07.txt

Search History:

User: /find API endpoints
BACR: [Opens search interface with 12 matches across 3 conversations]

API Reference

Authentication

BACR uses Google OAuth 2.0 for user authentication and server-based API key management.

RAG System

The RAG (Retrieval-Augmented Generation) system combines vector search, keyword search, and reranking.

Security

Data Storage:

Refresh tokens: Windows Registry (encrypted)
Conversations: Local SQLite (user-specific)
Documents: Local FAISS indexes (user-specific)
Preferences: Windows Registry (encrypted)

Data Transmission:

All API calls use HTTPS
Google OAuth uses PKCE flow
JWT tokens expire after 24 hours
Refresh tokens expire after 30 days

Privacy:

User data isolated by hash
No document content sent to server
Only embeddings stored locally
Token usage logged for billing only

Best Practices:

Logout when done to clear API keys
Don't share refresh tokens
Use ZDR mode for sensitive data
Regularly export important conversations

Additional Resources

Support

Email: support@bacr.app

Legal

Version: 1.0.0
Last Updated: March 7, 2026 Documentation: https://bacr.app/documentation

Maintained with care by The BACR Team