AICalcX — AI Project Cost Calculator

2025 | Python, Microsoft Agent Framework, Chainlit, Azure Retail API | GitHub

AICalcX is a real-time, agentic cost intelligence system powered by the Microsoft Agent Framework, built to compute and visualize AI project costs dynamically using live cloud pricing and autonomous reasoning.

Problem / Motivation

Organizations planning AI projects face critical challenges in cost estimation:

• Static Pricing: Traditional calculators use hardcoded prices that quickly become outdated
• Hidden Costs: Infrastructure, operations, and human resource costs are often overlooked
• Multi-Cloud Complexity: Difficult to compare costs across Azure, AWS, GCP, and other platforms
• Dynamic Workloads: Token consumption and resource usage are unpredictable
• Enterprise TCO: Hard to calculate accurate Total Cost of Ownership for large-scale AI deployments
• Transparency Issues: Black-box calculators provide no insight into cost breakdown methodology

AICalcX addresses these challenges with real-time pricing APIs, multi-agent orchestration, and transparent cost breakdowns that adapt dynamically to live Azure pricing, usage patterns, and AI model parameters.

Core Functionalities

Real-Time Pricing Intelligence

• Live Azure Pricing: Fetch current pricing from Azure Retail API with automatic updates
• Fallback Mechanisms: SerpAPI and cached JSON for when primary APIs are unavailable
• Cross-Cloud Ready: Azure support with upcoming AWS, GCP, Render, and Vercel integration
• No Hardcoded Prices: All pricing data sourced from live APIs or MCP-managed data

Multi-Agent Cost Analysis

• Intent Recognition: Natural language query parsing (e.g., "Cost of 1M GPT-4 tokens in eastus")
• AI Model Costs: Token-based pricing for GPT-4, embeddings, fine-tuning, and more
• Infrastructure Costs: Compute (VMs, containers), storage, and network bandwidth
• Human Resource Costs: Client-side and vendor-side personnel estimates
• Operational Overhead: Maintenance, monitoring, and contingency calculations
• Cost Aggregation: Unified reports with markdown tables and detailed breakdowns

Dynamic Scenario Simulation

• Token Computation: Understands shorthand notation (e.g., "500k", "2.5M", "1B")
• Usage Scaling: Model different user loads, agent counts, and token volumes
• Regional Pricing: Compare costs across Azure regions automatically
• SKU Variations: Handle multiple pricing tiers and subscription types

Transparent Reporting

• Detailed Breakdowns: Line-by-line cost calculations with formulas and references
• Export Options: Markdown, CSV, JSON, and PDF formats
• Live References: Links to Azure pricing sources for verification
• Analysis History: Local storage of previous cost analyses for comparison

Description / How It Works

1. User Query: Submit natural language query or structured parameters via Chainlit web interface
2. Intent Parsing: IntentAgent extracts model name, token count, region, and usage patterns
3. Pricing Fetch: PricingAgent retrieves live Azure pricing via Retail API or MCP cache
4. Fallback Handling: FallbackAgent uses SerpAPI or cached JSON if primary sources fail
5. Multi-Agent Analysis: Parallel execution of specialized agents:
6. AIAgent calculates model token costs
7. InfraAgent estimates compute and storage
8. HumanAgent computes personnel costs
9. OpsAgent adds operational overhead
10. Cost Aggregation: AggregatorAgent combines all outputs into unified markdown tables
11. Result Visualization: Display interactive cost breakdown with export options

Agent Architecture

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17

User Query
   ↓
IntentAgent → Extract model, tokens, region
   ↓
PricingAgent → Azure Retail API → Live pricing data
   ↓
FallbackAgent → SerpAPI / Cache → Backup pricing
   ↓
Parallel Agent Execution:
   ├─ AIAgent → Model token costs
   ├─ InfraAgent → Compute/storage/network
   ├─ HumanAgent → Personnel costs
   └─ OpsAgent → Operational overhead
   ↓
AggregatorAgent → Unified cost report
   ↓
Markdown / CSV / JSON Export

Each agent operates autonomously, fetching real-time data and updating a shared AgentContext before final aggregation.

Challenges & Issues Addressed

• Pricing Complexity: Multiple SKUs, dynamic regions, and unclear model mapping
• Azure API Gaps: Delayed updates and inconsistent SKU naming conventions
• Human Resource Variability: Non-standardized rates and team composition differences
• Usage Estimation: Unpredictable token consumption patterns across workloads
• Subscription Variations: Enterprise vs. retail pricing mismatches
• Cloud Type Differences: Public, government, and sovereign region pricing inconsistencies

Tech Stack & Frameworks

• Languages / Frameworks: Python 3.10+, Chainlit (Web UI)
• AI Orchestration: Microsoft Agent Framework (multi-agent coordination)
• Pricing APIs: Azure Retail API, SerpAPI (fallback)
• Data Persistence: GitHub MCP Server (version-controlled pricing cache)
• In-Memory State: Pandas DataFrames, Python dictionaries (shared agent context)
• Visualization: Plotly charts, Markdown tables
• Export Formats: Markdown, CSV, JSON, PDF

Features / Capabilities

• Live Pricing Fetch: Azure Retail API with SerpAPI fallback
• Multi-Agent Orchestration: 8 specialized agents for modular cost computation
• Natural Language Queries: Parse "Cost of 1M GPT-4 tokens in eastus" automatically
• Dynamic Token Computation: Handle shorthand notation (500k, 2.5M, 1B)
• Cross-Cloud Ready: Azure support with AWS/GCP/Render/Vercel upcoming
• Transparent Reports: Detailed cost breakdowns with live references
• Scenario Simulation: Scale users, agents, or tokens dynamically
• No Hardcoded Pricing: All data from APIs or MCP-managed sources
• Export Options: Markdown, CSV, JSON, PDF formats
• Analysis History: Local storage for tracking previous cost calculations

Example Query & Result

Prompt:

"Estimate cost of 1M GPT-4 input tokens and 500k output tokens in East US."

Result:

Additional infrastructure, human resource, and operational costs displayed in separate sections.

Potential Applications

• Enterprise AI Planning: Calculate accurate TCO before deploying large-scale AI systems
• Multi-Cloud Comparison: Compare costs across Azure, AWS, and GCP for cost optimization
• Budget Forecasting: Model different usage scenarios to predict monthly/annual costs
• Vendor Proposals: Generate detailed cost breakdowns for client proposals
• Cost Optimization: Identify expensive components and find cheaper alternatives
• Scenario Analysis: Test "what-if" scenarios for scaling AI workloads
• Compliance Reporting: Transparent cost calculations for audit and compliance

Future Enhancements

• Integrate AWS, GCP, Vercel, and Render pricing APIs
• Build interactive visualization dashboard with Plotly/Streamlit
• Add autonomous reasoning loop for cost optimization suggestions
• Implement budget alerting and cost tracking over time
• Develop REST/GraphQL API for programmatic access
• Add support for fine-tuning, batch inference, and GPU pricing
• Create browser extension for quick cost estimation

Learning Outcomes

• Built multi-agent orchestration system using Microsoft Agent Framework
• Integrated live cloud pricing APIs with fallback mechanisms
• Developed natural language query parsing for cost parameters
• Implemented modular agent architecture with specialized cost domains
• Designed shared agent context for coordinated multi-agent workflows
• Learned Azure Retail API nuances and pricing data structures
• Created transparent cost reporting with detailed formula breakdowns

Links

• GitHub Repository: AICalcX
• Known Blockers: BLOCKERS.md