Avaloka CLI
A team of data analysis agents that takes a dataset from raw rows to a trained model and checks its own conclusions.
- 122 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- The agent writes and runs Python code on your data locally or in a cluster and can train and deploy models.
Install
Manual install
git clone https://github.com/guruvaidev/avaloka.git && cd avalokaThe first quick start command. Then `./scripts/install.sh` and `source .venv/bin/activate`; installation takes 5–10 minutes.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent writes and runs Python code on your data locally or in a cluster and can train and deploy models.
This is third-party code. Review the repository files before installing.
What it does
Avaloka is an open-source CLI agent for working with data. The `avaloka analyze` command inspects a dataset, writes and validates the analysis code, runs it and prints a verdict, creating next to it a folder with an executive report, reproducible `analysis.py` code and a notebook, plus files that trace where each number came from. `avaloka train` fits several models, checks that the best one beats a trivial baseline and writes a deployment-ready model. `avaloka chat` talks through the data and `avaloka coordinate` runs several agent copies on parts of the data and merges them into one answer. Inside are twelve roles: planner, profiler, coder, validator, executor, trainer, visualizer and three checkers. They refuse to train on leaked targets, require beating the baseline on the confidence interval and check the report text against computed numbers. It runs on a laptop or a Ray and Kubernetes cluster, and the model can be Claude, GPT, Gemini or local vLLM and Ollama.
Who it is for. Analysts and data scientists who need a reproducible table breakdown with checked conclusions, not a chat answer.
Good fit when
- A new CSV must be analyzed: average bill per diagnosis, spread causes, data quality.
- A model is needed with proof that it beats a trivial baseline and did not train on target leakage.
- Analysis must be reproducible: code, notebook and the origin of every number are needed.
- A large file is split among several agent copies with `avaloka coordinate`.
Not a fit when
- You need a quick one-off answer from a small table: a plain model chat will do.
- You work on Windows: the project targets macOS and Linux.
- You need cloud database connectors and team features: they exist only in paid editions.
Example request
Analyze `healthcare_dataset.csv` with the goal average bill by condition and show the verdict and report folder.Limitations
Python 3.11 or 3.12 on macOS or Linux is required. The open edition is sized for a laptop or a small cluster, has not been tested on large multi-node clusters or production load, and support is community-driven. Cloud database connectors, cluster provisioning and team features exist only in commercial editions. The agent writes and runs Python code on your data, so run it on data and in an environment where that is acceptable. Without a model only the basic commands work, while conversation and rich narration need a key.
How to disable. Delete the repository clone and the `.venv` virtual environment. Analysis results and models sit in the `avaloka-analysis/` and `avaloka-model/` folders and are removed by hand.
FAQ
How do I run a first analysis?
Install with `./scripts/install.sh`, activate the environment and run `avaloka analyze app/sample_data/healthcare_dataset.csv --goal "average billing amount by medical condition"`. No API key is needed.
What is in the result folder?
The `executive_report.html` report, `README.md` with findings and limitations, `analysis.py` and `analysis.ipynb` code, the `goal_analysis.json` answer and the `lineage.json` and `data_quality.json` provenance files.
Which data formats are supported?
CSV, TSV, Parquet, Excel, JSON, XML, ORC and Avro.
Which model can I connect?
Claude through `ANTHROPIC_API_KEY`, OpenAI, Gemini on Vertex AI or a local one through vLLM or Ollama. The choice is set with the `INFERENCE_PROVIDER` variable.
What does `avaloka coordinate` do?
It clones the agent into several workers, each takes its part of the data and the main agent merges their profiles and verdict into one result in `avaloka-fleet/coordination.json`.
Discussion
No questions or reviews yet.
Related
CLI toolsA set of commands, skills and subagents for Claude Code that runs data analysis like a careful team analyst.
A template for data scientists: AGENTS.md, CLAUDE.md and skills for Python, SQL, Polars and model diagnostics.
A CLI agent for quantitative investment research: it reads broker reports, ideates factors and computes IC on offline data.
AVA is a visual-analytics skill that answers questions about a CSV or database with inspectable queries and a chart, not a guessed text answer.
A single binary indexes tables and logs so an agent gets a few matching records back instead of raw history
For a repeatable experiment, an agent downloads a Kaggle dataset to a working directory and retrieves notebook outputs.
README in short
The repository holds the CLI, a team of agents in Python, a web UI and API, a Helm chart for Kubernetes, bundled sample datasets, and documentation on installation, editions and architecture. The license is Apache-2.0, and the project is versioned with a changelog.