Pricing

Pay less for every insight

Credible sits between your AI and your warehouse, and meters in the same units they already do: tokens for AI, compute time for queries, and storage. Same meters, a fraction of the volume, a fraction of the cost.

Open Source

Free forever. Yours to run.

Local development and a server you host — your compute, your LLM, no meters:

  • Local-first development — model, query, and build data apps on your laptop
  • AI modeling and analysis agent — turns any coding agent into a data expert
  • Data apps and dashboards — one model, in your business or in your product
  • Materialization and pre-aggregation — sources, pre-aggregates, and indexes
  • Models, queries, dashboards as code — versioned in Git, shipped by CI/CD
View on GitHub

Cloud

Start free. Pay for what you use.

Everything in Open Source, managed by us — for one person or a whole company:

  • Fully managed — we run it all; nothing to deploy, patch, or scale
  • AI modeling and analysis agent, built in — the whole workflow in your browser, nothing to install
  • MCP for every agent — Claude, ChatGPT, Gemini, and your own agents draw the same governed context
  • Metered on tokens, queries, and storage — the same units your stack already bills in; see the rates below ↓

Enterprise

Built for scale. Priced for volume.

For BI and embedded analytics at scale. Everything in Cloud, plus:

  • Shared workspaces — analysis, apps, and dashboards in one governed home
  • Fine-grained access and audit logging — defined once, enforced everywhere
  • Forward Deployed Engineers — hands-on help from modeling to production
  • Dedicated clusters — multi-region, VPC connectivity, and Private Link
  • 99.99% SLA with 24/7 pager support — plus SOC 2 compliance

How the meters work

Three meters. Matched to your stack.

Every meter starts free: each month, every organization gets an allowance worth hundreds of questions and tens of dashboards on an average size database.

Tokens

matched to your LLM

The Credible app agent, Credible bot, and context retrieval all meter in tokens — counted the same way your LLM provider counts them, priced at the same rate as the frontier model that powers Credible.

$5.00
per million input tokens
$25.00
per million output tokens
Free each month
— 1M input + 250K output

Retrieval is metered on the tokens in the search request plus the tokens in the context returned. You pay the rate you already pay — the savings come from needing far fewer tokens to reach a correct answer.

Queries

matched to serverless DuckDB

Queries meter on compute time and bytes served — the seconds the serving layer spends answering, and the bytes it hands back.

$2.40
per compute hour, billed per second
$0.20
per GB served
Free each month
— 10 compute hours + 10 GB served

Compute time applies when Credible’s serving layer does the work, running the query on materialized sources. Queries that run directly on your warehouse incur no compute charge from Credible. Bytes served is egress — the same meter your cloud provider already runs whenever data leaves.

Storage

the cost of optimization

Capturing meaning is step one. Credible then optimizes your data and its meaning for serving — materialized sources, indexes, and models — and keeps them ready to serve, so retrieval and queries return without waiting on a warehouse scan.

$0.10
per GB / month
Free each month
— 1 GB storage

Storage is the optimization that makes the other two meters small: the materialized sources, indexes, and models that let retrieval and queries skip the warehouse.

Optimize your spend

Most platforms add to your bill. This one shrinks it.

The spend moves out of your LLM, your warehouse, and the legacy BI licenses you renew every year — and into your pocket. Credible's pricing captures a fraction of what you save; the rest stays with you.

There are three ways to put your data in front of your AI. All three reach the same answers. What differs is what you burn along the way — tokens on one meter, compute on the other.

The cheapest token is the one your AI never spends.The cheapest second of compute is the one it never runs.The cheapest BI license is the one you cancel.

Credible vs. Power BI, Tableau & LookerMigrating off your BI tool

The three ways
Way 1 of 3

Raw warehouse access

Your agent loads schemas and data samples into context, then relearns your business through trial and error — every single session. Long contexts re-read on every turn; wrong guesses regenerate. Both meters spin: token spend compounds with every exchange, and every sample, exploration, and retry is compute your warehouse bills for.

Way 2 of 3

Load the whole data model into context

Better — the meaning is written down. But your agent carries the entire model on every request, whether the question needs one measure or a thousand. Context bloats, accuracy degrades, retries multiply.

Way 3 of 3

The Credible Way

Your agent asks for what the question needs and gets exactly that: the right measures, dimensions, and business rules, access-controlled and ready to use. Underneath, the work is already done — the measures, dimensions, and joins defined; your data indexed, materialized, pre-aggregated, and cached, distilled down to what matters. Small requests in, small context out — that is the 10x on both meters: not cheaper rates, just far fewer tokens and far less compute to get a correct answer.

1/10
the tokens
1/10
the compute
You keep
the difference

Current estimates — an agent generating SQL against your raw lake, versus the same agent retrieving from the engine's concept index and warm storage sitting on top of it.

What's metered

Every meter, one table.

What each surface meters on, and what it costs. Anything marked free is free at any volume.

  • Credible app agent

    Tokens (input + output)

    $5.00 / M in · $25.00 / M out

  • Credible bot

    Tokens (input + output)

    Same token rates

  • Context retrieval

    Tokens (request + context returned)

    Same token rates

  • Claude, ChatGPT, Gemini & any agent via MCP

    Not metered

    Free — your agent, your tokens

  • Queries — Credible serving layer

    Compute time (billed per second)

    $2.40 / compute hour

  • Queries — direct on your warehouse

    Compute time

    Free — you pay your warehouse, not us

  • Query results — any source

    Bytes served (egress)

    $0.20 / GB

  • Storage

    GB stored (models, materialized sources, indexes)

    $0.10 / GB / month

  • Users & seats

    Not metered

    Free — unlimited

  • Model publishing

    Not metered

    Free

FAQ

Questions we get asked

The questions a pricing page owes straight answers to.

Why do you charge the same token rate as my LLM?
Because the savings should not come from a discount — they come from volume. Credible reaches the same answer with roughly a tenth of the tokens. Charging the rate you already pay keeps the comparison honest: same units, same rate, one-tenth the meter.
Where does the 1/10 come from?
Current estimates, comparing two paths to the same answer. The baseline is an agent generating SQL against your raw lake: it loads schemas, samples data, guesses, and retries — every exploration is tokens spent and compute burned. Credible’s path is the same agent retrieving from the engine’s concept index and its own warm storage over that same lake, where the measures, dimensions, and joins are already defined and the data is already aggregated and cached. Small request in, small context out. The ratio moves with your data and your questions; the direction does not.
What actually consumes tokens?
Three things: sessions with the Credible app agent (modeling and analysis), Credible bot conversations, and context retrieval — metered on the tokens in the search request plus the tokens in the context returned. Token counting works exactly the way your LLM provider counts them.
What about Claude, ChatGPT, Gemini, or my own agents connecting via MCP?
Free. Your agent brings its own LLM — its tokens are on your LLM bill, not ours. Credible charges nothing for MCP access; queries your agents trigger meter compute time and bytes served like any other query.
Why are queries billed on compute time, not bytes scanned?
Because it is the honest meter for the engine underneath. Credible’s serving layer runs on DuckDB over DuckLake, where the cost of a query is the seconds it takes, not the bytes it touches — so we meter the seconds, billed per second, at a rate matched to serverless DuckDB compute. Every second Credible spends replaces roughly ten your warehouse would have spent on the same question.
What is the difference between compute time and bytes served?
Compute time is the work of answering the query, and you are only charged when Credible’s serving layer does that work, running on materialized sources. When a query runs directly on your warehouse there is no compute charge from Credible; you pay your warehouse exactly as you do today. Bytes served is the result delivered to your agent, app, or dashboard — it maps to cloud egress, the meter your cloud provider already runs on every byte that leaves, and it meters the same way from either path.
What counts as storage?
The optimization. Storage is what Credible builds from your data and its meaning — materialized sources, indexes, and models — kept ready to serve so retrieval and queries return without waiting on a warehouse scan. It is the meter that makes the token and query meters small.
Do you charge per seat?
No. Credible meters like a utility: you pay for what flows through it, not for who is allowed to turn it on. Unlimited users on every plan.
What is the difference between Open Source and Cloud?
Where it runs and who runs it. Open Source is the engine on your machines: local development, a server you host, your own LLM, and no meters. Cloud is everything in Open Source, managed by us — we run the serving layer and the optimizations, you get the built-in agent and MCP for every agent, and the three meters above are the whole bill. It fits one person as well as a whole company; the meters are per organization, never per seat. The model format is the same, so a package you build locally publishes to either.
When do I need Enterprise?
When Credible becomes production infrastructure: an internal BI deployment for a large organization, mission-critical embedded analytics, or any need for dedicated infrastructure, shared workspaces, fine-grained access control, an SLA, or compliance.
Are the included amounts per user?
Per organization. All three meters — tokens, compute time, and storage — pool at the org level, and every user draws from the same pool. Adding teammates never changes your bill.

Ready when you are

Spend less on tokens. Burn less compute. Keep the difference.

Connect your warehouse, and watch what you don't spend.