albirint
In development — early access opening soon

Build verifiable data workflows

AI reasoning and planning. Deterministic actions. Full lineage.

Connect a spreadsheet, a cloud bucket, or a database, then ask a question in plain English. albirint returns a computed answer — tables, metrics and a written explanation — streamed live as the analysis runs.

Request early access — we’ll email you when accounts open
How it works

The AI plans. It does not do the math.

Most tools hand your question to a language model and print whatever it says. albirint splits the job in two, so you can check the work.

Reasoning

The model writes a plan

Your question goes to an AI planner, which produces an ordered list of steps — load this table, filter to Q1, group by region, sum revenue. The plan is visible before anything runs.

Execution

A deterministic engine runs it

Each step is a named action executed with standard data libraries — Polars, DuckDB and pandas. The same plan on the same data always produces the same numbers.

Lineage

Every step is recorded

Steps produce artifacts — stored intermediate results you can open and inspect. Artifacts expire; the recipe is kept, so an analysis can be re-run later on fresh data.

One question, start to finish:

What were my top-selling products last quarter by region?
Plan — written by the model

The model reads your question and decides which steps to run. It never computes the answer itself.

Executed deterministically
  1. 1load_tablesales.csv
  2. 2filter_rowsdate in Q1
  3. 3group_byregion, product
  4. 4aggregatesum(revenue)
  5. 5sortrevenue desc
Recorded

Every step stores its result. The notebook keeps the recipe, so the same analysis re-runs on fresh data.

Data sources

Bring your own data

Credentials are stored encrypted and are supplied to the analysis engine per request. Which source types a plan can connect is listed on the plans page.

File upload Available

CSV, Excel and Parquet files, uploaded straight into a notebook. Included on every plan.

SFTP Available

Pull files from an SFTP server on a path you control.

Backblaze B2 Available

Read objects from S3-compatible bucket storage.

PostgreSQL Available

Query an existing Postgres database over a read-only session.

MySQL Available

Query an existing MySQL database over a read-only session.

Warehouses In development

Snowflake, BigQuery, Redshift and Databricks are not available yet.

Plans

Same engine on every plan

Plans differ by weekly volume and which data sources you can connect. Full limits and answers to common questions are on the plans page.

Free

Try it on a single dataset.

$0
No card required
  • 20 queries per week
  • 1 notebook
  • File uploads up to 10 MB

Power

Heavy, repeated workloads.

$75/mo
$60/mo billed annually
  • 1,200 queries per week
  • 40 notebooks
  • 30 saved pipelines
  • Uploads up to 500 MB

Enterprise

Volume pricing and agreed terms.

Custom
Billed annually
  • Negotiated limits
  • Direct support contact

Early access

albirint is in development and not yet open for sign-up. Email us and we will contact you when accounts open. No charges are being taken today.

Email hello@albirint.com