AIHostBox
Client area Deploy an app
All 57 applications
CategoriesAI assistants & chatLLM runtimesAgents & RAGGateways & LLM opsVector & dataAutomationSpeech, image & videoSearch & web agentsSupport & CXML platformMCP servers
SolutionsPrivate team chatUnlimited automationCompany knowledge AIAI gateway & observability
More PricingGuides ComparisonsToolsPrivate LLM hosting LicensingSecurity SupportClient area

MLflow hosting

Experiment tracking and a model registry that outlive the notebook.

ML platformApache-2.0from $5.99/mo
In short

MLflow hosting at AIHostBox means a dedicated, single-tenant MLflow instance on our infrastructure: we install it, pin it to a reviewed release, put it behind TLS on your domain, back it up nightly and keep it patched. Plans start at $5.99 per month and the admin account, the data and any API keys are yours.

Run trackingModel registryArtefactsComparison UI

MLflow is the Apache-2.0 platform for tracking machine-learning work, started at Databricks and now developed by a large open-source community. Every training run logs its parameters, metrics and artefacts; the UI compares runs side by side; and the model registry gives each model versions, aliases and lineage back to the run that produced it. Autologging integrations for the common Python ML libraries mean adding it to existing code is often a line or two.

Recent releases, MLflow 3 in particular, widen the scope to generative AI: OpenTelemetry-based tracing of LLM applications and agents, evaluation with built-in metrics, and a prompt registry with versions and lineage. All of it works best as a tracking server that everyone points MLFLOW_TRACKING_URI at, rather than an mlruns folder on one laptop. A server gives the team one history, one registry and one place to look when a result from last quarter has to be reproduced.

MLflow hosting plans

StarterPersonal use & small teams
2 GB RAM · 1 vCPU · NVMe
$5.99/mo
Order
PlusBusy workflows & communities
4 GB RAM · 2 vCPU · NVMe
$9.99/mo
Order
ProHeavy pipelines & many users
8 GB RAM · 3 vCPU · NVMe
$16.99/mo
Order

Every plan is a dedicated single-tenant instance with NVMe storage, nightly off-box backups, TLS and managed patching. Quarterly, semi-annual and annual terms are discounted at checkout. Model tokens are not included — you bring your own key.

How we run MLflow

The MLflow tracking server (mlflow server) on port 5000 with Postgres as the backend store and a local artefact store proxied through the server, following the upstream self-hosting guide.

What we configure for you

  • --backend-store-uri pointing at the bundled Postgres
  • --artifacts-destination with artefact proxying, so clients need no storage credentials
  • Basic auth via --app-name basic-auth, with MLFLOW_FLASK_SERVER_SECRET_KEY and MLFLOW_AUTH_ADMIN_PASSWORD set per instance
  • --allowed-hosts set to your hostname for the DNS-rebinding protection
  • An S3-compatible artefact destination in your own bucket, on request

Which plan fits

Starter (2 GB) suits one person or a small team logging runs and a handful of registered models. Plus (4 GB) handles many concurrent jobs or steady LLM trace volume, and Pro (8 GB) is for heavy tracing and comparison views across thousands of runs. The server records results; the training itself runs on your own machines.

The honest limits

Built-in authentication is username and password with coarse permissions; there is no organisation-wide SSO in this setup. Large artefacts such as model weights fill local storage quickly, so bring an S3 bucket for serious volumes.

What people run MLflow for

Reproducible experiments

Know exactly which parameters produced a result.

Model lifecycle

Staging and production versions with an audit trail.

LLM tracing

Record the prompts, tool calls and responses of an agent as inspectable traces.

Prompt versioning

Keep prompts in the registry with versions instead of scattered through code.

Licence and version policy

ItemPosition
Upstream licenceApache-2.0
Sold as managed hostingYes
TenancySingle tenant — one customer per instance
PatchingCritical advisories within 24 hours, high severity within 72

Our full position on licences, including the applications we decline to sell, is on the licensing page.

Works well with MLflow

Jupyter logo

Jupyter

A persistent notebook server that does not sleep when your laptop does.

from $5.99/moDetails →JupyterHub logo

JupyterHub

Notebook servers for a whole team or classroom.

from $18.99/moDetails →Label Studio logo

Label Studio

Data labelling for text, image, audio and video in one tool.

from $11.99/moDetails →

More ML platform we host

Argilla logo

Argilla

Human feedback and dataset curation for LLM work.

from $11.99/moDetails →
Questions

MLflow hosting: your questions

Which database does a hosted MLflow use?
Postgres. MLflow’s backend store accepts PostgreSQL, MySQL, SQLite or SQL Server, and since 3.7 a bare install defaults to SQLite. For a shared server we use Postgres, so concurrent writes from many runs do not contend on a single file. Artefacts go to a separate artefact store.
How do people log in?
Through MLflow’s built-in basic authentication (--app-name basic-auth), which gives per-user accounts with READ, USE, EDIT and MANAGE permissions on experiments and models. We set MLFLOW_FLASK_SERVER_SECRET_KEY and the admin password; you create further users. Clients authenticate with MLFLOW_TRACKING_USERNAME and MLFLOW_TRACKING_PASSWORD.
How do artefacts get from my training jobs to the server?
The server proxies artefact uploads, so clients only need the tracking URI and their credentials — no storage keys on every machine. log_artifact and model logging send files through the server to its artefact store. If artefacts will be large, we can point that store at your own S3-compatible bucket.
Can I migrate runs from my existing MLflow server?
From a SQL backend, send a database dump and your artefact directory; we restore both and run mlflow db upgrade if versions differ, keeping run IDs and registry history. From a file-based mlruns folder, the open-source mlflow-export-import tool copies experiments, runs and registered models between servers.
Can MLflow serve my models for inference?
MLflow can package a registered model and serve it with mlflow models serve, but these plans are sized for tracking and the registry, not for inference traffic. Most teams run the tracking server here and deploy models from the registry to their own serving infrastructure.

Run MLflow without running a server

A dedicated MLflow instance, deployed and maintained by us. Month to month, cancel whenever.

Browse the catalogue Talk to a human