FAQ
Frequently asked questions about Credible
Straight answers about Credible, the AI Analytics Engine: what it is, how it works, what data it connects to, which AI tools use it, what it costs, and how it keeps data secure.
The basics
What Credible is
- What is Credible?
- Credible is the AI Analytics Engine. Connect the data you have, tell Credible what it means, and it turns your data into trustworthy answers, apps, and agents. The engine generates everything else: the pipelines, the storage, and the layer that serves every answer. No warehouse, no pipelines, and no data team are required. Every agent and app answers from the same definitions, so the numbers agree.How Credible works →
- Who is Credible for?
- Credible is built for three groups. Small teams: companies with no data team, or a team of one, who want answers, apps, and agents with the people they already have. Builders: AI and software engineers shipping agents and apps who need context they can trust, not hallucinations they have to debug. Data teams: the people who own what the data means and want to be the foundation, not the bottleneck.
- Do I need a data team to use Credible?
- No. Credible is designed so a company without a data team can get trustworthy answers. An agent drafts the data model from your answers using open-source modeling skills, you confirm what is right, and the engine generates the rest. Teams that do have data engineers use Credible too, in their own IDE and workflow.
- What is an AI Analytics Engine?
- An AI Analytics Engine takes a data model as input (what your data means: the metrics, entities, joins, and rules of your domain) and generates the infrastructure that delivers it. In Credible that means the semantic layer that serves consistent answers to every agent and app, plus the pipelines, materialized storage, and indexes underneath. You own the what. The engine generates the how.
How it works
From your data to answers
- How does Credible work?
- In three steps. First, connect the data you have: databases, spreadsheets, CSVs, or a warehouse if you have one. Second, tell Credible what your data means: an agent drafts a data model from your answers with open-source modeling skills, and you confirm what is right. Third, ask questions and build on the answers: every question goes through one gateway, is checked against who is allowed to see what, and is answered from the same definitions.Read the introduction →
- How long does it take to get started?
- About five minutes to a first answer. Sign up free, connect your data, and ask a question in the browser with nothing to install. Developers can start from a coding agent instead with the cred CLI or the VS Code extension.Model locally with a coding agent →
- What is Malloy, and why does Credible use it?
- Malloy is an open-source data modeling language created by Lloyd Tabb, the co-founder of Looker and creator of LookML. It is MIT-licensed and a Linux Foundation project. Credible data models are written in Malloy because it is a real language rather than YAML: joins are declared in one place, joined totals never double-count, definitions compose, and it compiles to SQL for BigQuery, Snowflake, Postgres, Databricks, DuckDB, and more. AI agents read and write it fluently.Why Malloy →
- Do I have to write Malloy myself?
- No. In Credible an agent drafts the data model with open-source modeling skills while you answer questions about what your data means in plain language. You review and approve what it proposes. Engineers who prefer to edit the model directly can, in the browser, in VS Code or Cursor, or with a coding agent such as Claude Code.Build a model with AI →
- How does Credible keep AI agents from making up answers?
- By giving the agent your definitions instead of raw tables. When you publish a model, Credible compresses it into a concept index: every source, measure, and dimension, its documentation, and the indexed values. When an agent asks a question, Credible returns just the slice of the model that answers it, with queries grounded in governed views. The agent works from what your metrics mean rather than guessing at column names.How agents get context →
Your data
What Credible connects to
- What data sources does Credible support?
- Credible connects to BigQuery, Snowflake, PostgreSQL (including Supabase, Amazon RDS, Cloud SQL, Azure Database, and Neon), MySQL, Databricks, Trino, DuckDB, MotherDuck, and DuckLake. Through DuckDB it also reads CSV, Parquet, and JSON files in place in Amazon S3, Google Cloud Storage, or Azure Data Lake. Spreadsheets (.xlsx), CSV, and Parquet files can also be uploaded with a model.Connect your data →
- Do I need a data warehouse to use Credible?
- No. Credible brings its own query engine and storage, so a warehouse is optional, and so is a database. You can start from spreadsheets, CSVs, or files in object storage. If you do have a warehouse, Credible works with it and can run queries directly against it.
- Can Credible connect to Shopify, Stripe, or HubSpot?
- Not directly today. Credible connects to databases, warehouses, spreadsheets, and flat files. If your data lives in a SaaS app, export it to CSV or a spreadsheet, or sync it into a database you already use, and connect Credible to that.
- Does Credible copy or store my data?
- Only the parts you choose. By default Credible queries each source live. You can opt parts of a model into persistence with an annotation, and Credible then keeps managed, derived copies, such as materialized tables, rollups, and indexes, so queries return without a warehouse scan. Copies that are no longer used are cleaned up. Connection credentials stay in Credible and never sit on a laptop.Persistence and materialization →
Agents and apps
Where the answers go
- Which AI assistants work with Credible?
- Any agent that speaks MCP. Credible connects to Claude (including Cowork, through a connector or plugin), ChatGPT (in developer mode), and Gemini Enterprise, and to coding agents including Claude Code, Codex, Cursor, VS Code Copilot, Gemini CLI, Windsurf, and opencode. Your own agents can connect over MCP or the REST APIs, and Credible bot answers questions in Slack.Connect your AI assistant →
- Does Credible have an MCP server?
- Yes. Credible serves every data model over MCP, the open protocol that connects agents to tools. An agent calls get_context to retrieve the part of the model relevant to a question and execute_query to run a governed query. Access is authenticated with OAuth or an API key, and every query is checked against the caller’s permissions.Credible MCP server →
- Can I build data apps on Credible?
- Yes. A data app in Credible is HTML and JavaScript that ships with a data model and queries it through the same governed gateway as every other surface. Credible serves it to signed-in users, enforces access per query, and can isolate each of your own customers’ data with row-level rules. Most teams build apps by describing them to an agent.Data apps →
- Can I embed Credible in my own product?
- Yes. Credible exposes a Data API for running queries, a Retrieval API for retrieving model context, and an Admin API for managing environments, packages, connections, and permissions. Data apps and AI features in your product query the same data model as your internal agents, so customer-facing numbers match internal ones.Credible APIs →
- Does Credible do dashboards?
- Yes. Describe the dashboard you want and the agent builds it from your data model, with KPI cards, charts, tables, and filters. Because dashboards read from the same definitions as your agents and apps, every surface shows the same numbers.Analytics →
Pricing
What it costs
- Is there a free plan?
- Yes, two. Credible Cloud starts free: every organization gets 1 million input and 250,000 output tokens, 10 compute hours, and 1 GB of storage each month at no charge, with no credit card or sales call. The Open Source plan is free forever: run Malloy Publisher on your own machines with your own LLM and no meters.Pricing →
- How is Credible priced?
- By usage, like cloud infrastructure. Credible Cloud meters three things: tokens ($4 per million input, $20 per million output), query compute ($2.40 per compute hour, billed per second), and storage ($0.10 per GB per month). Users, model publishing, and MCP access are free. Enterprise adds dedicated infrastructure and support, priced for volume.See all rates →
- Does it cost extra to connect Claude or ChatGPT?
- No. MCP access is free, and your agent brings its own LLM, so its tokens stay on your LLM bill. What the agent does through Credible meters like anything else: context retrieval in tokens, queries in compute time.
- When do I need Credible Enterprise?
- When you need dedicated single-tenant clusters (multi-region, VPC, Private Link), fine-grained access control and audit logging, a 99.99% SLA with 24/7 support, or Forward Deployed Engineers working alongside your team.Compare plans →
Security
Trust and governance
- Is Credible SOC 2 compliant?
- Yes. Credible is SOC 2 compliant, covering the Security trust services criteria. Customer data is encrypted in transit with TLS 1.2 or higher and at rest with AES-256. Request the security package, including the auditor’s report, from security@credibledata.com.Security at Credible →
- Does Credible train AI models on my data?
- No. Under the Credible Platform Agreement, Credible will not train any AI model on customer data without your prior written consent.Platform Agreement →
- How does Credible control who sees what?
- At one gateway, on every query, whether it comes from an agent, an app, or a dashboard. Environment roles (Admin, Modeler, Viewer) control who can model, publish, and query. On Enterprise, access rules written in the data model also restrict which sources, rows, and columns each user or customer can see, and every query is recorded in an audit log.Security and permissions →
- Can I self-host Credible?
- Credible Cloud is fully managed. If you want to run on your own infrastructure, Malloy Publisher, the open-source server Credible is built on, serves the same data models over REST and MCP. Enterprise customers who need isolation or data residency get dedicated single-tenant clusters on AWS or Google Cloud.Deployment options →
Comparisons
Credible and the alternatives
- How is Credible different from a BI tool like Looker or Power BI?
- A BI tool is built around dashboards for a central data team. Credible is an engine: it serves one data model to every agent, app, and dashboard, charges for what you use, and does not need a warehouse or a data team to start. Your definitions are written in open-source Malloy, so they are not locked to one vendor’s tool.Credible vs. Looker →
- Is Credible a semantic layer?
- No. Credible is an engine that generates a semantic layer. A semantic layer is a tier you install between storage and your tools, then keep in sync by hand. Credible takes your data model as input and generates that tier, along with the pipelines and storage beneath it, and runs all of it. The data model is the part you own; the semantic layer is output.Malloy and the semantic layer →
- How is Credible different from dbt, Cube, or another semantic layer?
- dbt is the transformation standard, and Credible works on top of the tables it builds. The dbt Semantic Layer and Cube are semantic layers: a tier you configure in YAML and operate between your warehouse and your tools. Credible is the engine that generates the semantic layer from a data model written in Malloy, along with the pipelines and storage beneath it. Existing dbt, Cube, LookML, and Omni definitions are a starting point for the model, not a rewrite.All comparisons →
- Why not just point an AI agent at my database?
- Because an agent writing SQL against raw tables has to guess what each column means, and every guess costs tokens and compute. Two people asking the same question can get two different numbers. With Credible the agent retrieves your definitions from the concept index and queries governed views, so it reaches a correct answer faster and every answer agrees.Credible vs. building it yourself →
- Can I migrate from Looker, Tableau, or Power BI?
- Yes. Credible has migration guides for Looker, Power BI, Tableau, Omni, Sigma, Cube, dbt, Snowflake, and Databricks. An agent reads your existing models and dashboards as prior art and drafts the Malloy data model from them.Migration guides →
Company
Who builds Credible
- Who makes Credible?
- Credible Data Inc., based in Boulder, Colorado. Founder and CEO Kyle Nesbit spent seventeen years at Google building Google Cloud’s data analytics stack, including bringing Looker into Google Cloud and Malloy into BigQuery. Credible raised a $10 million seed round in July 2026 from Gradient, SignalFire, and K5 Global.About Credible →
- Which companies use Credible?
- Customers include bars.com, which cut campaign reporting from a week to minutes without a data team, and VideoAmp, which took definition coverage from 47% to 100% across more than 2,000 definitions.Case studies →
- Is Credible open source?
- Credible is open core. Malloy, the modeling language, is MIT-licensed under the Linux Foundation. Malloy Publisher, the server that serves Malloy models over REST and MCP, and the agent skills for modeling and analysis are open source and maintained by Credible. The managed engine, Credible Cloud, is a commercial service.Malloy Publisher on GitHub →
Still have a question?
Try it on your own data
Connect your data, tell Credible what it means, and ask. Start free. No credit card, no sales call.