fullOpen.fullExtensible.

An open-source no-code Visual ML platform. No paywall, no closed components, no external API keys.

View on GitHub
localhost:8000
EN
dash.AI
DatasetsModelsGenerativePlugins

Welcome to dashAI!

Pick a module to start

Datasets
IMPORTCLEANEDA
Models
TABULARNLPREGR
Generative
LLMT2IINFER
Plugins
INSTALLEXTEND
Manifesto

An open platform isn't one distributed for free: it's one that behaves openly.

dashAI is Full Open + Full Extensible. That's the editorial line.

[ 01 ] Openness
{ I }Four proofs

Full Open.

Openness isn't a license statement: it's a measurable property of the system.

01 / License

MIT licensed, no copyleft.

Integrable into institutional systems and commercial products. No obligation to release what you build on top.

02 / Core

No productive paywall.

Scheduling, deployment, RBAC and governance live in the core. The project you clone today is what reaches production.

03 / Ecosystem

No closed components.

Every official extension shares the core license. Auditable, modifiable, forkable under the same terms.

Native coverage, local execution.

⊳ Openness, in practice
Tabular

Classification & regression

15 + 15

Thirty models over scikit-learn and boosting. Each one as a subclass of BaseModel.

  • RandomForest
  • XGBoost
  • LightGBM
  • SVM
  • MLP
  • +25
NLP + Translation

Transformers & seq-to-seq

15 + 9

Classifiers and neural translation models with native BLEU · ChrF · TER metrics.

  • DistilBERT
  • DeBERTa-v3
  • ModernBERT
  • NLLB
  • OpusMT
  • +19
Generative models

Local LLMs

5

Llama, Mistral, Mixtral, Qwen and SmolLM on open weights. Generation as a native task.

  • Llama
  • Mistral
  • Mixtral
  • Qwen
  • SmolLM
Generative vision

Text to image

11

PixArt-Sigma, SDXL, Stable Diffusion and ControlNet variants. Automatic CUDA detection.

  • PixArt-Sigma
  • SDXL
  • SD v3
  • SD v2
  • ControlNet
  • +6
[ 02 ] The application
{ ■ }Schema-driven UI

The interface that emerges from the schema.

When you register a component, the UI isn't programmed: it's generated. The server exposes its Pydantic schema and the React frontend renders the form without prior knowledge.

localhost:8000 / models / tabular-classification
Models / Tabular / Configure

RandomForestClassifier

Hyperparameters generated from MyForestSchema

n_estimatorsint · 10–1000
3501000

Number of trees in the forest.

max_depthint | None
None

Maximum depth of the tree.

criterionenum
gini

Function to measure split quality.

bootstrapbool
TrueON

Use bootstrap samples when building each tree.

Form auto-generated from schema·4 fields

1. Schema → UI

Each form field is born from a Pydantic line. Type, range, default and description render automatically.

2. No frontend code

The new component touches zero React files. The UI shows up on its own once the backend registers it.

3. Browsable catalog

Tabular, NLP, translation, LLMs and generative vision coexist in one consistent UI — all on the same twelve abstractions.

Catalog / Models
[ TABULAR · 30 ]
Pick a model
RandomForest
Random forest
XGBoost
Gradient boosting
LightGBM
Microsoft GBM
SVM
Support Vector
MLPClassifier
Neural net
KNN
k-Nearest
Results / Job #218
[ COMPLETED · 2m 14s ]
Validation metrics
accuracy0.9342
f1_macro0.9118
precision0.8987
recall0.8512
Jobs · Queue
[ 8 BaseJob · 2 running ]
Orchestration
Runningtrain_rf_v268%
Runningexplain_shap12s
Donepredict_test1m 4s
Queuedoptimize_hpo
[ 03 ] Extensibility
{ II }Twelve abstractions

Full Extensible.

Each functional role has its base class. Twenty to forty lines for a new component. The UI emerges from the schema — it isn't programmed.

01
BaseTask
Task
Extensible paradigms: classification, regression, NLP, translation, generation.
02
BaseModel
Predictive model
Random Forest to transformers. Subclass + schema, ready for HPO.
03
BaseGenerativeModel
Generative model
LLMs and image as first-class citizens.
04
BaseMetric
Metric
Classification, Regression, Translation. Auto-appears in UI.
05
BaseOptimizer
HPO optimizer
Optuna + HyperOpt. Bayesian, evolutionary, QMC strategies.
06
BaseDataLoader
Data loader
CSV, JSON, images, audio. Each loader declares its format.
07
BaseExplorer
EDA / exploration
Visualizations, profiling, distributions — typed.
08
BaseConverter
Converter
Typed transformations between formats and representations.
09
BaseGlobalExplainer
Global XAI
Permutation importance, surrogate, partial dependence.
10
BaseLocalExplainer
Local XAI
SHAP, LIME, counterfactuals with dedicated abstraction.
11
BaseJob
Orchestration
Train, predict, explain, explore, convert, generate, evaluate, optimize.
12
BaseGenerativeTask
Generative task
Conditional generation, text-to-image, seq-to-seq.

Schema-driven UI

The server exposes the Pydantic schema; the React frontend renders the form without prior knowledge.

Hot-install from the UI

One click installs an extension from PyPI. No restart, no external shell.

Public HTTP API

Any HTTP client — Python, Jupyter, CLI — queries components and launches jobs. GET /component/, POST /job/.

FastAPI + React, not desktop

The server runs anywhere, reachable from any browser — local, network or remote.

my_classifier.py
# A complete classifier in dashAI
from dashai.base import BaseModel
from dashai.core.schema_fields import schema_field, optimizer_int_field
from sklearn.ensemble import RandomForestClassifier

class MyForestSchema(BaseSchema):
    n_estimators: schema_field(
        optimizer_int_field(ge=1),
        placeholder={
            "optimize": False, "fixed_value": 100,
            "lower_bound": 50, "upper_bound": 200,
        },
        description="Number of trees",
    )
    max_depth: schema_field(
        optimizer_int_field(ge=1),
        placeholder={
            "optimize": False, "fixed_value": 2,
            "lower_bound": 2, "upper_bound": 10,
        },
        description="Maximum depth",
    )

class MyForest(BaseModel):
    SCHEMA = MyForestSchema

    def train(self, X, y):
        self.model = RandomForestClassifier(
            **self.params
        ).fit(X, y)

    def predict(self, X):
        return self.model.predict(X)
33 lines · UI auto-generatedpython ≥ 3.10
[ 04 ] Download

Download dashAI.

v0.9.3-alpha·MIT License·Linux · macOS · Windows·May 2026
01Executables (recommended)

Native installer

No Python, no setup. Download the executable for your OS from GitHub Releases.

View releases
02PyPI

Install with pip

dashai starts the server and opens the browser at localhost:8000.

$ pip install dashai
$ dashai
# http://localhost:8000/ — Browser opens on its own.
View on PyPI
Requirements

What you need

  • Python≥ 3.10Only for pip install
  • RAM8 GB16 GB for local LLMs
  • OSLinux · macOS · WindowsNative executables available
Want to modify the core? Clone the repo:git clone github.com/DashAISoftware/DashAI
[ 05 ] Community

Let's build this together.

dashAI is open source and open community. Join the conversation, propose components, share experiments.

Contact

Reach out directly.

Questions, collaborations, institutional integrations, or just to share what you're building with dashAI.

dashai.nocode@gmail.com →
Discord

Join the Discord server.

Live support, release announcements, threads per module (Datasets, Models, Generative, Plugins) and a #show-and-tell channel to share your experiments.

Join Discord
Newsletter

One monthly digest.

New features, models added to the catalog, community plugins, relevant papers. No spam, no promos — just what matters for your workflow.

· Unsubscribe anytime · 0 commercial emails

[ 06 ] Collaborators

Built across institutions.

dashAI is a Chilean academic project maintained by a network of research institutions and public funding bodies.