Databox CLI - Your agentic analytics platform in the terminal
by•
Create data sources, load data, build metrics, and automate workflows directly from your terminal. Use 90+ commands yourself, or let AI coding agents like Claude Code, Cursor, and Codex do the work with 10 ready-to-use agent skills.
Replies
I'm a big fan of CLIs and use them a lot. They're really not as scary as they might sound :) And now Databox (finally) has one too!
What I also like is that anyone can create their own API key and use the CLI, regardless of their Databox plan or user role. Permissions still follow your existing role, so there's nothing extra to manage.
This makes life a lot easier, especially when you're working with scripts, automations, and agents.
@mateja_verlic_bruncic Thanks Mateja! Agree, CLIs are a lot friendlier than people expect, especially once an agent is doing the typing. And yes, anyone can create their own API key now. Since the key acts as you, it follows your existing role and access. No separate permission setup, which keeps scripts and agents simple and safe.
A must-have in the AI/Agentic era.
cracked team
@mcarmonas Thanks Marti!
Another step toward making analytics easier for AI agents to set up, manage, and automate!
@gregor_sulcer Thanks Gregor! That’s the direction - less clicking through setup screens, more letting agents handle the routine work.
Boy do I like not leaving my terminal these days. MCPs are cool, but I like spamming my terminal commands as in the old days more ;)
@aljaazm Ha, same here Aljaz. Both have their place - MCP when you want to chat with your data inside Claude or ChatGPT, the CLI when you want to script it, pipe it, or let an agent run it in the terminal.
When you envision the ultimate user for this product. What does he/he use it for, and how does he/she use it? What problem do they have, and how does this solve it and what does the outcome look like? Trying to understand it a bit more. Thanks
@alchemist_studio Great question, André. The ideal user is someone technical enough to work in a terminal or with an AI coding agent like Claude Code - a developer, a data or ops person, or a technical marketer at a SaaS company or agency.
The problem: a lot of important data doesn’t come from a ready-made integration. Product data from your own database, CSV exports, internal tools. Getting it into Databox used to mean manual uploads or writing custom code against the API. Agencies have a second problem: setting up the same data sources, datasets and metrics for every new client by hand.
How they use it: install the CLI, add the skills, and tell the agent what you want in plain words. “Create a dataset for our monthly revenue export and load this file.” The agent picks the commands, sets up the dataset, loads the data and checks it worked.
The outcome: data that was stuck in files or scripts shows up in Databox on its own, next to everything else, with metrics and dashboards on top. Repeat work turns into one command or one prompt, and the team spends its time on the numbers instead of moving them around.
This fits how I actually work — most of my data starts in a script, not a dashboard, so piping JSON straight into a dataset from a nightly job is the part that caught me. That's the first thing I'd hand over to it. With 91 commands mirroring the API, do the agent skills end up being how most people actually use it day to day, or do you still find yourself reaching for the commands directly?
@jcbuilds Both, and they split nicely. The skills shine for setup and one-off work: creating the dataset, defining the schema, building metrics, or digging into something you haven’t touched in a while. Once the pipeline is set, the nightly job itself is best as a plain command. No agent needed, same result every run.
For your case that looks like: let the agent set up the dataset once, then have the script run databox dataset ingest <id> --file output.json each night with DATABOX_API_KEY set. Two things to keep in mind: batches cap at 500 records, and adding --idempotency-key makes a rerun after a failure safe.
gotta share this with my builder friends asap
@shushantlakhyani Thank you! Happy to hear the feedback
alright, i'm all about making my terminal life easier. honestly, the thought of not having to switch tabs to check analytics feels like a dream. but 91 commands? yeah, i might just cherry-pick the ones i actually learn.
@therayess Cherry-picking is the right way to use it. Most people only need a handful of commands day to day, usually dataset ingest, metric list and metric drilldown. The skills are split by area too, so you can install just the ones you use, e.g. npx skills add databox/databox-cli --skill databox-metrics.
If Claude Code pushes a dataset schema change through the CLI while ingestion already has rows in the old shape, does it reject the change or migrate the existing data?
@igorblink Good question. Neither - the CLI doesn’t change a dataset’s stored schema after the first ingest. dataset update only renames the dataset. Changes like renaming columns, converting types, adding formula columns or hiding columns go through modifications, which sit on top of the stored rows. Claude Code can run dataset preview-modification first to see the result before saving anything. If new records don’t match the schema, the ingestion fails and the error names each rejected record, so nothing gets silently mixed. For a real breaking change, the clean path is a new dataset and a re-ingest.