Experiment tracking
Runs, parameters and results kept in one place.
AI Development
MLflow development services for AI you can audit
We have no published MLflow build. Our AI product keeps an audit trail of every brief and reports cost and tokens per workspace, which is the problem MLflow also addresses.
0Published builds on MLflow
3.18.0Latest release, 2 Nov 2026
5Releases since 17 Jun 2026
1Audit-trail build nearby, Sales OS
What we build
What we would take on with MLflow
Tracing
Each step of an AI call recorded so a failure can be found.
Evaluation
Scores that carry the evidence next to them, as in Sales OS.
Cost per workspace
Provider cost and token counts reported for each team.
Architecture
Where MLflow would sit
MLflow beside an AI product
App
React 19 appAccount briefs
API
FastAPIPython 3.13
Claude-compatible APISales OS, today
Tracking
MLflowWhere it would sit
Audit trailID, workspace, date, engine version
Illustration. A dashed outline marks where the technology would sit. We have no published build on it.
Releases
MLflow ships often
The latest MLflow releases
31 Jul 2026
MLflow 3.15.0
Released.
3 Sep 2026
MLflow 3.16.0
Build your own trace UI by describing it.
6 Oct 2026
MLflow 3.17.0
Typed decisions in evaluation and tracing.
2 Nov 2026
MLflow 3.18.0
Automatic issue detection, an agent skill registry and better evaluations.
Source: MLflow releases page, mlflow.org/releases, checked 5 November 2026. The site calls MLflow the leading open source AI engineering platform.
Builds
The closest builds we can show you
AI product, in-house
Sales OS
Every dossier keeps its ID, workspace, date and engine version, and the tool reports cost and tokens per workspace.
40 accountsAbout 26 hours of reading a week replaced by 40 credits
Logistics, fleet operations
TruckFleet Labs
A Python 3.13 back end with a sync queue, where an outage lost no data.
71Queued records delivered automatically
Fit check
When MLflow is the right call
Pick MLflow when
- Several people train and compare models
- You need to explain an AI answer later
- You run your own model services
Look elsewhere when
- You need a published MLflow build to inspect, which we do not have
- One model that never changes
- A hosted API already shows your usage
Explore
Same job, other tools
Pairs with
Where we used it
Stack
What MLflow usually sits beside
Python
PyTorch
Apache Airflow
Docker
scikit-learn
TensorFlowFAQs
What buyers ask first
Which MLflow version is current?
MLflow 3.18.0, released on 2 November 2026. The previous releases were 3.17.0 on 6 October, 3.16.0 on 3 September and 3.15.0 on 31 July.
Do you have a published MLflow build?
No. We say so on the first call and show the nearest work, which is Sales OS, with an audit trail and cost reporting per workspace.
What does MLflow 3 focus on?
The release notes stress generative AI and agent work: tracing, evaluation and issue detection.
Who owns the code and the cloud accounts?
You do, from the first commit. Repositories and cloud accounts stay in your name.
Tell us what your AI has to prove
Thirty minutes on your models, your evaluations and your costs. No sales pitch.