• 

  CostGraph UI

CostGraph

The true cost of one business outcome — across shared infrastructure, cross-project services, and AI.

The problem

Every engineering team can tell you their total cloud bill — but not what one order, one transaction, or one AI decision actually costs to produce. That cost is smeared across shared infrastructure, services owned by other teams, and AI/LLM spend. The gap drives bad pricing, capacity, and optimization decisions every day.

What CostGraph does

CostGraph computes the fully-loaded cost of one business outcome by attributing shared cloud cost, cross-project service cost, and AI cost down to the unit. It shows why cost changed, simulates optimizations that respect quality and prod guardrails, and includes a natural-language agent that calls real deterministic engines and proves its work with a visible tool trace. The story is KNOW → EXPLAIN → OPTIMIZE.

How it uses AI

1. It measures agentic workflows — costing a multi-step agent pipeline (fraud-triage → payments → support) per successful outcome, exposing what retries and failures actually cost.
2. It is agent-operated — a natural-language agent interprets questions, orchestrates the cost engines, and explains results, but never computes a number itself. Deterministic core, AI on top: the math is auditable; every answer shows the engines it called.

What's built

• Adoptable connector framework — four normalized data planes
• Working connectors — GCP billing (CSV + BigQuery) and OpenAI usage, plus file; cloud-agnostic
• Five deterministic engines — attribution, unit economics, AI tokenomics, RCA, constrained optimization
• AI agent endpoint — NL → real engine calls → grounded answer with tool trace
• React dashboard — cost decomposition, allocation control, interactive cost-flow graph, tokenomics, agentic-workflow tracing, optimization simulator
• 13 tests, Apache-2.0, public repo

Attribution uses an OpenTelemetry-baggage model for cross-project cost. Baggage values are seeded in this build to demonstrate the model; live OTel collection is the documented integration. The demo uses seeded, plausible figures so it runs standalone — the allocation logic and attribution model are real; only the input data is seeded.

Differentiation

FinOps tools tackle cloud cost allocation. CostGraph goes further by unifying cloud + cross-project + AI cost per successful outcome in one graph — and by making the allocation methodology an interactive control.

Built with

Python · FastAPI · Pydantic · React · Vite · Tailwind · OpenAI · GCP BigQuery · OpenTelemetry

Links

• GitHub: https://github.com/sgade123/costgraph
• Demo video: https://youtu.be/j2KfdzFiRLo
• Presentation deck: https://docs.google.com/presentation/d/1V6lVlkOE3kkboB2EDPOhd7IK2jGs25wsqhSj8h5IezE/edit?usp=sharing

Built With

• fastapi
• gcp-bigquery
• openai
• opentelemetry
• python
• react