Costex Independent product · 2026
From cloud spend
to a decision.
I designed and built Costex to carry a cost question from the daily cloud picture through a project change, a savings review and a monthly Excel workbook. The core runs privately in a customer-hosted workspace.
Try the workspace
The workspace
Five areas. One workspace.
Overview and Explore share date and cloud controls. Explore follows cost development and project changes through to the charges behind them. Optimize keeps current findings; Exports uses a billing month. Cost-centre assignments stay in Settings.
Overview
Where is the spend?
Daily cloud costs, provider trends and potential monthly savings.
Explore
What changed?
Project cost development, the three largest increases and decreases, and scoped charge detail.
Optimize
What should we review?
A current queue grouped by action, with financial context and evidence on demand.
Plan
What comes next?
Budgets, forecasts, commitments and unit economics.
Exports
What goes to finance?
Monthly tenant workbooks, native Excel PivotTables, saved files and close evidence.
Agent, wherever you work. Analyze costs, find savings or prepare Excel workbooks from the conversation canvas or the compact floating window. Supporting sources open on demand.
The through line
One question can cross the whole bill.
Costex keeps the evidence and the decision together, without making a missing source look like a clean result.
Find the charge behind the change.
Start with the three largest project increases and decreases, then open the exact project or charge detail. The selected dates and cloud stay in scope. Missing billing history is marked unknown instead of shown as zero.
Review the useful savings together.
Optimization findings are grouped by action across clouds. Conditional prices, observed costs and unknowns remain distinct. Ignore records a reason, and overlapping alternatives are counted once in the potential.
Prepare the month without rewriting history.
Choose a billing month, review each tenant and create native Excel workbooks in one batch. Persistent cost-centre assignments carry forward; saved files and close evidence remain available when needed.
Design principle
Clarity includes
the limits.
Source before conclusion. Billing, inventory and utilization have different coverage.
Facts before estimates. Forecasts and provider savings suggestions are labelled as such.
Preview before write. Scope, affected objects and revisions are checked before a change is saved.
How it is built
A normalized billing layer supports AWS, Azure, Google Cloud and Open Telekom Cloud. BigQuery is the warehouse path; operational decisions and revision history live in a separate transactional store. Python produces the native Excel workbooks. The same scoped actions serve the interface and Agent.
What the public demo proves
The public workspace runs the same interface with isolated fictional data. Visitors can explore cloud spend, review grouped opportunities and generate sample Excel workbooks. The sample Agent uses prepared answers and bounded synthetic tools, with no AI provider or cloud credentials.
Live provider feeds, inventory, customer identity, invoice evidence and external delivery still need configuration and acceptance in each deployment. A prepared change is not a cloud mutation, and an estimate is not verified savings. The product source is private.