AI writes the code.
You see what it’s worth.
Cost tracking for teams that build with AI. Sessions, commits, CI, deployments and cloud — in one chain from prompt to production.
NewRecommendations with evidenceThe real Nythrex app, recorded on demo data from the Atlas example.
- GitHub
- GitLab
- Bitbucket
- Azure DevOps
- Jira
- Trello
- Linear
- Azure
- Google Cloud
- Claude Code
- Codex
- GitHub Copilot
Connects to the tools you already use
A new kind of tool for engineering spend. Nythrex links what AI spent with what the team delivered — so every figure can be explained and every recommendation checked.
Metadata only
A developer machine sends time, model, tokens and the commit. Never code or prompts.
Rules, not generation
Every recommendation is a named rule over your records. The same data gives the same conclusions.
Unknown ≠ zero
If a source did not report a price, the total is marked incomplete. The gap is shown, not hidden.
1.0 · Sources
Sources
and collection
One authorization per source and one command per machine for AI. Repositories and tasks stay in your tools — records arrive on their own.
Capabilities
GitHub
Commits, pull requests, CI runs and deployments — the work and the CI of one repository.
Azure DevOps
Boards, Repos and Pipelines: tasks, code and CI/CD from one connection.
Microsoft Azure and Google Cloud
Cloud spend by subscription, project and service — at project level, not split across changes.
GitHub Copilot
Copilot usage of your organizations, and your personal usage and billing by model.
GitLab and Bitbucket
Merge and pull requests, commits and pipelines — the same chain as GitHub, from one authorization.
Jira, Trello and Linear
Tasks and status transitions. Time from task to production is counted from here.
Claude Code and Codex
One command per machine configures the tools it finds. For a team, one settings block.
Coming soon
AWS and Cursor. These sources are not available yet — they show where the product is heading.
2.0 · Change chain
The cost of every change,
link by link
Task, AI session, commit, pull request, CI and deploy become one unit of work. Click the links — or remove CI and see what a gap looks like.
Atlas / Checkout
Example project · demo data
STEP 1 OF 7 · TASK
ATL-118 · Checkout address validation
Jira · picked up at 09:12
Time to production is counted from here.
Known cost of this change
$0.00
- AIreported by source—
- CImeasured × price—
- Cloud · Atlas project
- $42.00 · not split across changes
- People’s time
- estimate · optional
Capabilities
Exact links
The commit names the AI session; the pull request names the commit. Nothing is matched by guessing from timestamps.
The basis of every figure
Reported by the source, measured × price or estimated — every number carries its label.
Cloud at project level
The cloud bill is not split artificially across changes: it sits next to them, on its own line.
The state of every unit
In progress, merged, delivered or abandoned — from source events, not from a hunch.
Visible gaps
No CI runs means the link is marked as a gap and the total as incomplete. Nothing is filled in.
People’s time — optional
The cost-of-time estimate can stay off. A member sees only their own.
3.0 · Overview
Overview
and outliers
Five seconds to understand, three clicks to prove it. See how the cost of delivered work changes, and what stands out.
Capabilities
Three numbers on the first screen
Spent, delivered, could be saved — and the change against the previous period.
Project, person, team, period
The same records in any cut — no separate reports.
Unknown price — kept apart
Incomplete totals are labelled, and AI with no project stays on its own line.
Only the outliers
Not everything at once, only what stands out. Every line leads to where it can be examined.
Evidence on the side
Records, the basis of a figure and its arithmetic open in a side panel, without leaving the page.
Weekly digest
The week’s highlights by email — optional, with one-click unsubscribe.
4.0 · Recommendations
Recommendations
with evidence
A dashboard shows. Nythrex also tells you what to change: rules find where money produced no result and show the arithmetic.
AIAlready spent15 sessions
AI sessions with no commit$37.00
5 of 15 finished AI sessions produced no commit — half of the AI spend for the period.
How it was calculated
- A session counts as finished after a few hours without activity.
- Small sessions below the threshold are not counted.
- This is an upper bound: research and reading code land here too.
What you can do. Check whether long sessions end in a commit and whether the hook that signs commits with the session is installed.
The rules are real; the examples are demo data.
Capabilities
Rules, not generation
A recommendation is the result of a named rule over your records. Every number has its arithmetic.
Did it work?
Mark a recommendation as fixed — after at least a week you see the number before and after.
AI · CI · Cloud · Process
Sessions with no commit, an expensive model on helper requests, redundant workflows, cloud with no owner, abandoned work.
Evidence or silence
Too few records — the rule stays silent and says how many more it needs.
Potential ≠ actual
Possible savings are never added to what has already been saved.
Nothing to run
Rules fire on their own after every data refresh.
Your code stays yours. Our numbers stay honest.
Code, prompts and file names never leave the developer’s machine.
Metadata only
An empty spot is never passed off as zero spend.
Unknown ≠ zero
The product does not compare people or rate productivity.
Analytics ≠ surveillance
What arrives from a developer machine
- time, model and token count of every request
- the cost Claude Code estimates itself
- session ID, account email and organization
- repository URL, branch and commit of the session
What is never stored
- prompt and response text
- code and file contents
- file names and tool parameters
Codex adds tool arguments and output snippets to its telemetry and does not let you turn this off. Nythrex drops them on arrival, without reading or storing them.
Before you start
Honest answers
to awkward questions
What happens after sign-up?
Three short steps: your tools, your focus and how you found us. Then your workspace. Your own data appears once sources are connected; until then the workspace honestly shows an empty state.
Do you see our code or prompts?
No. A developer machine sends the time, model and token count of a request, its cost, the session ID, the repository URL, branch and commit. Prompt and response text, code, file contents and file names are not stored.
Is this a tool for monitoring developers?
No. The product does not compare people with each other and does not rate productivity. The cost-of-time estimate is optional; a member sees only their own, and the owner and admins see the overall picture.
How accurate are the costs?
Nythrex tells apart amounts reported by a source, calculations from a price list and estimates — the basis is shown on every figure. An unknown price never becomes zero: the total is marked incomplete. Cloud is counted at project level and is not split across changes.
Where do recommendations come from? Are they AI-generated?
No. Every recommendation is the result of a named rule over your records: the same data always gives the same recommendations. It shows the records it is based on and the arithmetic behind its amount. With too few records, the rule stays silent and says how many more it needs.
Do we have to move our work into Nythrex?
No. Repositories and tasks stay in your tools. Nythrex collects records through integrations and refreshes them automatically — there is nothing to run.
Which AI tools are supported?
Claude Code and Codex. Each machine connects with one command that configures the tools it finds; for a team there is one settings block.
Are there paid plans?
Paid plans and billing have not launched yet. Creating an account does not start a subscription and needs no card.
Stop guessing.
Start seeing.
No card and no subscription: paid plans have not launched yet.