MinusX suits founders and small data teams that want dashboards, alerts and plain-English answers before hiring an analyst. Self-hosting costs nothing and the cloud plans offer a trial. Treat the vendor's number-one benchmark claim with care: the public leaderboard now shows its entry at 66.0%, below the leading submissions.
What changed
MinusX is an open source agentic business intelligence platform from a two-founder San Francisco team in Y Combinator's Summer 2024 batch. Its agent answers questions, builds dashboards, sends alerts and works through Slack or an MCP (Model Context Protocol) connection. It reads sources such as PostgreSQL, BigQuery and ClickHouse, and runs self-hosted or in the vendor cloud.
Maker: MinusX · Autonomy: semi autonomous · Maturity: GA
Underlying models: Models from OpenAI, Anthropic and Google (vendor-selected, or your own API keys)
About MinusX
MinusX is a San Francisco company from Y Combinator's Summer 2024 batch, founded by Vivek Aithal, a former comma.ai research engineer, and Sreejith Puthanpurayil, who worked on ads delivery at Meta. Its GitHub description calls the product "Claude Code for data": an agentic business intelligence platform where the agent, not the human, is the main user. It counts as an agent rather than a tool because the platform exposes questions, dashboards and reports as files the agent can read and write, runs scheduled jobs that watch metrics, and investigates the cause when something breaks.
The hard part of data work is context rather than SQL, and the product is built around that idea. A shared knowledge base stores definitions such as what ARR or an active user means in your company, learns from each conversation, and can be tested with evals. People reach the agent in the browser, in Slack or through a Model Context Protocol (MCP) connection, which lets other AI clients query the same data. The repository lists sources including PostgreSQL, BigQuery, ClickHouse and MySQL, plus CSV files and Google Sheets. The model layer comes from OpenAI, Anthropic and Google, chosen by the vendor or supplied through your own keys.
The evidence deserves a careful read. On DataAgentBench, a Berkeley benchmark of data tasks, the homepage claims the top spot as of June 2026, but the public leaderboard read in October 2026 lists the MinusX entry among mid-table submissions, behind several newer ones. The product began as a Chrome extension for Metabase and Jupyter in 2024, then moved to its own open source platform, whose repository dates from February 2026. The community is small, so expect to lean on the docs and the vendor.
Self-hosting is free, and managed cloud plans start at a per-seat price listed under pricing. It suits startup founders and small data teams with a question backlog. Larger organisations that need governed enterprise BI with embedded analytics should look at ThoughtSpot or Sisense first, while text-to-SQL products such as BlazeSQL and AI2sql cover the narrower job of writing queries.
Other products in the same area include Tellius for guided insights, Akkio for forecasting, Qlik for established BI, Databricks for the warehouse side, Secoda for data catalogs and Rows for spreadsheet analysis. If you are choosing between an agent-led BI shell and a classic dashboard tool, Smart Match can compare them against your brief.
Pricing
MinusX lists four plans. Open Source costs nothing to self-host with your own LLM keys. Founders is $40 per user per month with 500 agent credits per user and no platform fee. Team is $600 per month with 20 users included and 10k pooled credits. Enterprise is custom-quoted, typically above 20 users. The page shows an annual option marked as saving 20% next to a monthly one, so confirm which billing cycle the displayed figures reflect. Credit top-ups, storage and a data modeling service are marked coming soon. A 7-day trial applies to the cloud plans. For published per-seat prices elsewhere, compare [ThoughtSpot](/hub/tools/thoughtspot) or [Akkio](/hub/tools/akkio).
Key Features
- Knowledge base that learns: Metric definitions such as ARR or active users are stored in a shared knowledge base, and the homepage shows entries tagged as auto-learned from earlier conversations.
- Evals for the agent: Teams write test questions and see each graded pass or partial, as in the vendor demo where two of three tests pass and one partially matches.
- Proactive alerts: Scheduled jobs watch metrics and notify you when something breaks, and agents then investigate the root cause across dashboards and data, per the repository README.
- Slack bot and MCP: A Slack bot and an MCP integration let people query data from chat, or from any other MCP client, without opening the BI interface.
- Dashboards, reports and decks: The agent drafts dashboards, reports, slide decks and scrollable data stories end to end, and can modify existing questions and dashboards.
- Ten data source types: The repository architecture notes list DuckDB, BigQuery, PostgreSQL, SQLite, MySQL, Athena, MongoDB, ClickHouse, CSV files and Google Sheets.
- One-command self-hosting: A curl installer needs Docker, then walks through the workspace, an AI model with a live connection test and an optional first database.
- Works with or without dbt: The README says it is built to run on messy SQL as well as modeled data, since few fast-growing companies have everything modeled.
Strengths
- The free self-hosted edition publishes its source on GitHub, so a team can inspect the code and keep the warehouse inside its own environment.
- The Trust Center lists both SOC 2 Type II and Type I as compliant, with the Type II report available on request for the cloud edition.
- Development is active: the latest repository commit landed in October 2026, and the history holds more than 1,000 commits since February 2026.
- Bring-your-own model keys let a team pick its LLM provider and pay that provider directly instead of buying credits.
Weaknesses
- The vendor's number-one DataAgentBench claim dates from June 2026, yet the leaderboard now shows its entry at 66.0% Pass@1 over five trials, behind newer submissions led by 86.2%.
- Licence wording is inconsistent: GitHub shows AGPL-3.0, while the pricing FAQ says Apache 2.0. Check the LICENSE file before embedding or modifying it.
- The community is small, with 44 GitHub stars, 4 forks and a two-person team on its Y Combinator profile, so support depth outside paid plans is untested.
- Credit top-ups and storage add-ons are marked coming soon, so a heavy user on a cloud plan cannot yet buy extra credits.
Frequently Asked Questions
How much do you pay for MinusX?
Self-hosting the open source edition is free. The managed Founders plan for solo founders and small teams runs $40 monthly per seat, and Team runs $600 monthly for a block of 20 users who share a credit pool. Enterprise is quoted individually. The vendor says bringing your own LLM keys takes 50% off any plan, and its annual option saves 20%.
Is MinusX free to use?
Yes, the open source edition is free forever according to the pricing page. You install it with Docker, supply your own model keys and pay the model provider directly. Cloud plans start with a 7-day trial and, per the vendor, no credit card. Self-hosters carry the upkeep, and the AGPL-3.0 licence shown on GitHub carries copyleft duties if you modify it and offer it as a service.
What should you use instead of MinusX?
ThoughtSpot fits larger organisations that want a full BI rollout with enterprise support. BlazeSQL is a managed text-to-SQL product with published monthly plans, and Tellius offers guided insights on custom pricing. Akkio adds forecasting and model training on its higher plan.
Is MinusX better than ThoughtSpot?
For a founder or small team, MinusX is the lighter start: you can self-host it free, and the agent builds the first dashboards for you. ThoughtSpot has the longer enterprise track record and embedded analytics, with sales-led contracts at the top end. The deciding factor is team size and governance needs, not raw answer quality, since neither vendor offers an independent head-to-head test.
How long does it take to get going with MinusX?
The vendor's demo video, titled Setup in seconds, shows Google Sheets turned into a dashboard in about five minutes. Self-hosting needs Docker and one install command, and the script asks for a workspace, an AI model and optionally a first database. Those timings are the vendor's own, not independently measured.
Top Alternatives
- ThoughtSpot: Pick ThoughtSpot for governed enterprise BI with embedded analytics; pick MinusX if you are a founder or small team that wants an open source agent first.
- BlazeSQL: Pick BlazeSQL for a paid cloud product with published plan prices and no free plan; pick MinusX to try a free self-hosted edition first.
- Tellius: Pick Tellius when a sales-led, custom-priced platform suits your procurement; pick MinusX for listed prices and an open source option.
- Akkio: Pick Akkio if forecasting and model training matter most; pick MinusX for dashboards, alerts and a learned knowledge base.