Keboola

Editor’s note: This Keboola review is based on my hands-on test of the Free plan in October 2026, from signup to a scheduled flow. I weighed pricing and credit consumption, connectors and CDC, transformations and orchestration, AI features, and the trade-offs against Fivetran and Airbyte.

Quick verdict: I recommend Keboola for small data teams and consultancies that want ingestion, SQL/Python transformations, scheduling, and Streamlit apps together. Standard-project credits vary by workload and backend size; Free’s up to 60 monthly minutes leaves little room for daily schedules.

Keboola, formerly Keboola Connection, is based in Prague. In this Keboola review, I’ll take a closer look at pricing, connectors, transformations, AI tools, and governance to help you decide between its bundled managed ELT platform and a dedicated ingestion service. Start with Keboola‘s no-card Free plan.

Key Takeaways

• Free costs $0 without a card: One project includes 120 job minutes in month one, then up to 60 monthly, capped by the previous month’s usage.
• Top-ups cost $0.14 per minute: Unlimited pipelines do not mean unlimited compute. Enterprise requires a custom quote.
• Standard-project rates vary sharply: SQL consumes 6 to 26 credits per hour; Python consumes 0.2 to 2.
• Standard-project users are all administrators: Governed roles require Governed Change Management.
• AI consumption is metered: Free includes the MIT-licensed Keboola MCP Server. Kai requires organization administrator activation and bills per reply.

Pros and Cons

How Much Does Keboola Cost?

Keboola has 2 plans: Free at $0 and Enterprise at a custom price. On Free, 1 credit equals 1 hour of job time from any component; standard contract projects use different credit rates by workload and backend size.

• Free ($0): For building and testing a small pipeline without a credit card. It includes 120 free job minutes in month one, then up to 60 monthly, with each month’s allowance capped at the previous month’s usage. Unused free minutes do not carry over.
• Compute and storage: One project, SQL and Python transformations, an extra-small Snowflake backend, and 250 GB of storage.
• Included tools: Unlimited pipelines, Flow Builder, and Keboola MCP Server.
• Restrictions: No Data Catalog, no time travel, and best-effort support without an SLA.
• Top-ups: Pay $0.14 per minute by card. Purchased minutes never expire while the project stays active and are non-refundable.
• Enterprise (custom pricing): For teams needing governed development, tailored deployment, and dedicated support.
• Collaboration: Data Share/Catalog, Data Apps, and Workflow Templates.
• Deployment: Tailored CDC and streaming, plus Dev/Prod mode with Git CI/CD and SOX controls.
• Hosting and compliance: Public or private SaaS in a VPC, with SOC 2 Type II, GDPR, and HIPAA coverage.
• Access and service: SAML plus Google/Azure SSO, storage options including Snowflake, BigQuery, and DuckDB, and a dedicated technical account manager.

Plan	Price	Included compute	Projects	Transformations	Key extras
Free	$0; $0.14/minute top-ups	120 minutes initially; then up to 60/month	1	SQL, Python	250 GB, Flow Builder, MCP Server
Enterprise	Custom quote	Contract-specific credits	Contract-specific	Broader engines and backends	Catalog, governance, deployment options, dedicated support

The following rates apply to standard contract projects; Free counts plain job minutes. A Small SQL job consumes 6 credits per hour on a standard project, versus Free’s 1 credit per job hour.

Workload	Size or basis	Credit consumption
Data source job	Per job hour	2
Data destination job	Per GB out	0.2
SQL job	Small / Medium / Large, per hour	6 / 12 / 26
Python, R, DuckDB	XSmall / Small / Medium / Large, per hour	0.2 / 0.4 / 0.6 / 2
dbt	Small / Remote, per hour	6 / 2
Streamlit app	XSmall / Small / Medium / Large, per hour awake	0.1 / 0.2 / 0.5 / 1
Python/JS app	XSmall / Small / Medium / Large, per hour awake	0.3 / 0.6 / 1.2 / 2.4
Kai	Median conversation; billed per reply	About 1.1; 300-credit one-time allowance per new organization

Python/JS app rates took effect October 15, 2026; they previously matched Streamlit. CDC costs $0.17 per minute or starts at $1,300 monthly flat, plus $700 monthly per replication slot on Enterprise.

Is Keboola Good Value for Money?

• Free top-ups equal $8.40 per job hour: Budget against component runtime, not pipeline count. Adding another pipeline configuration does not itself consume the allowance.
• Weekly schedules suit Free better: The tutorial flow takes about 4 minutes. Weekly execution fits within 60 minutes; daily execution at that size consumes about 120 minutes over 30 days, before development runs.
• Python has cheaper standard-project rates: Python and DuckDB consume 0.2 to 2 credits per hour, against SQL’s 6 to 26. That rate advantage does not apply to Free’s plain-minute billing.
• Rivals meter different units: Fivetran Free includes 500,000 MAR, Airbyte Standard starts at $20 monthly, and Hevo Free includes 1 million events.

Reviewer’s Notes: I recommend Free for building and testing, then Enterprise when the Catalog, governed roles, or tailored CDC become requirements. For daily schedules, budget top-ups from the start: 60 monthly minutes cannot cover daily runs of the tutorial flow.

My Experience With Keboola

Signing Up and Creating My Free Project

I signed up without entering a credit card and created my Free project. Keboola offered one project without a hosting-stack choice, so I did not select a cloud or region during setup.

I used the dashboard’s task-specific links to move through the pipeline:

• Components: I set up the HTTP extraction.
• Transformations: I created the Snowflake SQL job.
• Storage: I inspected buckets and tables.
• Conditional Flows: I connected extraction and transformation.
• Apps: I accessed data app creation.
• Jobs: I checked execution details.

The top bar included the Kai Agent button, but an organization administrator had to enable Kai.

Loading Data With the HTTP Extractor

I opened Components, searched for HTTP, and selected the plain HTTP connector, not HTTP Advanced Beta. I used the Boolabean cafe tutorial dataset:

• Sales records
• Staffing records
• Weather files
• Forecast files

1. Create the configuration: I selected Connect To My Data, named the configuration, and clicked Create Configuration.
2. Set the endpoint: I switched to Advanced Mode and entered the Base URL with its trailing slash. The Advanced Mode switch was one-way.
3. Map the files: I added a Row for each file; each row name became its table name.
4. Replace existing contents: I left Incremental load off, then selected Run Component.

I opened Storage to inspect the configuration’s bucket with its IN badge. The HTTP-loaded columns arrived as text, so extraction did not remove the need for SQL data preparation.

Writing a SQL Transformation

I opened Transformations, selected Create Transformation, and chose Snowflake. I added Table Input mappings with names matching the SQL, then used New Table Output for the result.

I selected Create Multiple Queries, entered the SQL, saved it, and clicked Run Transformation. I checked the job’s Success badge and opened the output in Storage.

The cafe dataset produced staffing_outlook with 42 rows and 10 columns, stored in an out bucket named after the transformation.

Scheduling a Flow

I opened Conditional Flows, selected Create Flow, and chose Build It Yourself. I placed extractors together in phase 1 and the transformation in phase 2, keeping the SQL job behind its required inputs.

1. Test execution: I saved the configuration and selected Run flow.
2. Set recurrence: In Schedules, I chose weekly Monday at 06:00 and retained UTC.
3. Add failure alerts: In Notifications, I added a recipient to the Errors card.

I checked execution through All Runs and Jobs. Flows defaulted to 3 retries with a 10-second delay.

Reviewer’s Notes: I recommend the weekly schedule for Free: the tutorial flow takes about 4 minutes, while daily execution exceeds 60 monthly minutes. Parallel tasks shorten elapsed flow time, but they do not reduce credit consumption because every component job bills separately.

Connectors and Change Data Capture

Keboola advertises 1,500+ data sources; the component catalog lists about 84 data source connectors and 26 destinations, plus Generic Extractor and Generic Writer for unsupported APIs.

Unlisted or private components can be added by ID, without functionality guarantees; experimental components are public code. Fivetran offers 700+ managed connectors and is my stronger choice when packaged ingestion coverage takes priority over Keboola’s transformation and app layers.

Keboola’s CDC uses Debezium, with database-specific setup requirements:

• Self-serve databases: MySQL 5.7+, MariaDB 11.1.2+, and PostgreSQL 10+.
• On-request databases: Oracle, MongoDB, SQL Server, and Cassandra.
• PostgreSQL restrictions: The connector remains in public beta and requires wal_level=logical plus a replication slot.
• Added columns: New columns appear, but historical records have empty values for them.
• Dropped columns: Existing destination columns remain and receive NULLs.
• Renamed columns: Manual remapping is required.

PostgreSQL snapshots resume at table granularity, and Append mode can duplicate rows. Its default snapshot batch contains 10,240 rows, and the heartbeat defaults to 3,000 ms.

On August 6, 2026, Google Ads data source jobs failed for 2 hours 46 minutes with UNSUPPORTED_VERSION after an upstream API version retired; some jobs may have written extra columns. On standard projects, a data source job consumes 2 credits per job hour and a data destination job consumes 0.2 credits per GB out.

Estuary Flow combines real-time CDC and batch ingestion, while Rivery publishes log-based CDC billing at 2 credits per 100 MB. Both belong on a replication-focused shortlist alongside Keboola.

Best for: I recommend Keboola for mixed API and database pipelines with transformations attached. Skip if: Choose another option when mature self-serve PostgreSQL CDC is non-negotiable.

Transformations, Flows and Dev Branches

Transformations

I recommend Keboola over a separate dbt-and-Airflow setup when a small team needs SQL transformations and scheduling inside one project. Free stops at SQL and Python; broader engine support belongs to subscription projects.

• SQL: Snowflake or BigQuery transformations.
• Code: Python on Free; subscriptions add R, Julia, optional Spark, and MLflow.
• Additional engines: DuckDB and dbt jobs.

Python/R/DuckDB backends range from XSmall with 8 GB RAM and 1 shared CPU to Large with 114 GB and 14 dedicated CPUs. Large also includes 1 TB SSD; standard-project credit consumption rises with backend size.

Flows

Keboola runs phases sequentially and tasks within a phase in parallel. I used that separation in the cafe pipeline to keep transformation behind extraction without maintaining an external scheduler.

• Schedules: Every 15 minutes, hourly, daily, weekly, end of month, or custom.
• Event triggers: Run a flow when a Storage table changes.
• Conditional execution: Only the first matching condition executes.
• Concurrency: The recommended maximum is 10 concurrent component jobs.

Order conditions by priority: a later matching condition does not create another execution path.

Dev Branches and Roles

Branched Storage isolates development buckets and tables from production on Snowflake and BigQuery. Rollout across all stacks is scheduled from November 2, 2026, with projects using processed tags or queries excluded for now.

Branched Storage is a prerequisite for Branches 2.0 merge requests. The current kbagent CLI supports GitOps pull/diff/push; legacy kbc remains supported.

• Listed roles: Share, Admin, Guest, and ReadOnly.
• Governed roles: Developer, Reviewer, and Production Manager require Governed Change Management.
• Standard-project access: Every user in Users & Settings is an administrator.

Finer access in standard projects requires splitting projects or using Storage Tokens. Matillion suits warehouse-native visual transformations, while Fivetran’s dbt integration suits ingestion-first stacks.

AI Features: Kai, the MCP Server and Data Apps

Kai

Kai can write SQL/Python and read job logs, but an organization administrator must enable it through the Kai Agent button. Keboola warns: “Kai is an AI and can make mistakes”.

• Pipeline assistance: Integration setup and troubleshooting.
• Modeling: Data models and semantic-layer work.
• Team memory: Rules taught to Kai apply to teammates after administrator approval, available since October 7, 2026.

Kai bills per reply, with about 1.1 credits per median conversation, not a fixed conversation price. New organizations receive a one-time 300-credit allowance shared across projects.

Keboola MCP Server

The MIT-licensed Keboola MCP Server launched June 13, 2025; version 1.110.6 arrived October 30, 2026. It exposes about 45 tools.

• Data operations: Storage inspection, SQL queries, and component configuration.
• Execution: Jobs and flows.
• Application work: Data Apps, streams, and the semantic layer.
• Clients: Claude Desktop, Claude Code, Cursor, Windsurf, and Make.

I opened Project Settings, then MCP Server, and used the project MCP URL for Claude Code:

claude mcp add --transport http keboola MCP_URL

I authenticated through OAuth and selected the project. Token mode uses an access token plus project ID for containers and CI.

Since October 23, 2026, query_data is read-only; changing tables requires a SQL transformation. Other tools still change the project:

• run_job
• modify_flow
• update_sql_transformation

Data Apps

Keboola supports Streamlit and Python/JS apps. I opened Apps and entered a prompt for a draft with a live preview; the draft remained “Not live yet” until publication.

Publishing targets Production with Basic (Password) authentication by default. Apps sleep after inactivity and consume credits only while awake, including idle time before sleep.

Reviewer’s Notes: I recommend allowing read tools while requiring approval for each update or run action. Test changes in a development branch first; the read-only restriction on query_data does not make the whole MCP connection read-only.

Security, Hosting and Reliability

Enterprise includes SOC 2 Type II, GDPR, and HIPAA coverage; Free provides no hosting-stack choice.

• Paid US hosting: AWS Virginia and GCP Virginia.
• Paid EU hosting: AWS Frankfurt, Azure Ireland, and GCP Frankfurt.
• Encryption: AES-256 at rest and TLS 1.2+ in transit.
• Authentication: TOTP and FIDO/U2F MFA, plus SAML SSO.
• Data exit: A take-out button in project settings and API export.

Keboola’s status page logged operational interruptions in September and October 2026:

• October 21 to 22: Medium and Large Snowflake transformations ran on the Small warehouse for about 32 hours.
• October 21 to 22: File uploads were disabled for about 27 hours after a vulnerability patch, affecting UI uploads, Kai chat attachments, and synchronous multipart imports. There was no evidence of exploitation.
• September 28: AWS EU flows were delayed for 5 hours 51 minutes, while individual jobs remained unaffected.

I recommend Enterprise support commitments and explicit recovery procedures for deadline-sensitive pipelines. Free’s best-effort support has no SLA, making it a poor fit for that workload.

How Does Keboola Compare to Competitors?

I favor Keboola when bundled Streamlit apps and Kai matter alongside ELT; I favor the alternatives below when ingestion or deployment control matters more.

• Fivetran: Free includes 500,000 MAR, with 700+ managed connectors across the platform; Standard offers 15-minute syncs and Enterprise 1-minute syncs. Paid standard connections add a $5 base charge to MAR billing. I favor Fivetran for hands-off ingestion at scale.
• Airbyte: Standard starts at $20 monthly, Plus at $189, and the 14-day trial includes 400 credits. I choose Airbyte Core for free self-hosting; Cloud Standard syncs hourly, while Plus supports 15-minute syncs and includes 40 credits.
• Hevo Data: Free includes 1 million monthly events and 5 users, while Starter costs $299 monthly and Professional $849. I favor Hevo for no-code pipelines with event-based budgeting, but Free limits connector access and schedules runs hourly.
• Matillion: Developer, Teams, and Scale charge credits per task hour, without a public dollar price per credit. I favor Matillion for warehouse-native visual transformation; validation and sampling are not billed, but a quote is necessary to calculate dollar costs.
• Rivery: Base costs $0.90 per credit, log-based CDC consumes 2 credits per 100 MB, and Base syncs every 60 minutes. I favor Rivery for published CDC billing math, with Base limited to 1 environment and 2 users.
• Estuary Flow: Estuary combines real-time ELT with CDC and batch ingestion. I shortlist it when replication drives the purchase and Keboola’s transformation and application layers are not priorities.

Keboola bills Free projects by job minutes, Fivetran by MAR, and Hevo by events. Compare the workload each allowance covers; a larger headline number does not establish a cheaper pipeline.

Tool	Free option	Paid entry	Billing unit
Keboola	120 initial minutes; then up to 60/month	$0.14/minute top-ups; Enterprise quote	Job minutes or workload credits
Fivetran	500,000 MAR	Usage-based plus connection charges	MAR
Airbyte	Self-hosted Core; Cloud trial	Standard from $20/month	Volume or credits, depending on plan
Hevo Data	1 million events/month	Starter $299/month	Events
Rivery	14 days or 1,000 credits	Base $0.90/credit	Credits

How I Reviewed Keboola

I signed up for Free and built the cafe tutorial pipeline with an HTTP extractor, Snowflake SQL transformation, and weekly flow. I followed the output through Storage and checked execution through All Runs and Jobs.

I also went through Kai access, the Keboola MCP Server setup, and data app creation. I separated Free’s job-minute billing from standard-project credit rates and weighed Enterprise governance against the administrator access in standard projects.

I compared Keboola with Fivetran, Airbyte, Hevo Data, and Matillion on ingestion coverage and transformation depth. My recommendation also weighed PostgreSQL CDC’s beta status, AI write permissions, and the September and October service interruptions.

Prices current as of October 2026.

Should You Use Keboola for Managed ELT?

I recommend Keboola for small data teams and consultancies that need ingestion, transformations, and scheduling in one platform. Its SQL/Python workflow and AI tooling offer more than a connector service, but Free is a testing allowance, not a generous daily production budget.

For consultants, extra Free projects for different client companies are available through support. The default account includes one project, so that support route matters when separating client work.

I favor Fivetran when hands-off, high-volume ingestion is the priority, and Airbyte when self-hosting is essential. I recommend skipping a standard Keboola project when fine-grained user roles are non-negotiable: every listed user is an administrator, and governed roles require additional capabilities.

As the pricing breakdown shows, component runtime and backend size determine the bill, not pipeline count. Start a Free project and run the tutorial flow before talking to sales, using its job consumption to judge your intended production schedule.

FAQs

Is Keboola free?

Keboola has a $0 Free plan without a credit card requirement. It includes 120 job minutes in month one, then up to 60 monthly, capped at the previous month’s usage; unused free minutes do not roll over. Extra Free projects for different companies are available through support.

How much does Keboola cost per minute?

Free-plan top-ups cost $0.14 per minute, equivalent to $8.40 per job hour. CDC costs $0.17 per minute or starts at $1,300 monthly flat, plus $700 monthly per Enterprise replication slot. Standard-project credits follow workload-specific rates, and Enterprise requires a custom quote without a public dollar price per credit.

What happens when my Free minutes run out?

Jobs pause when Free minutes run out, but the data stays. If the project remains inactive for 2 months after minutes run out, it faces suspension and eventual deletion. Purchased minutes are non-refundable but never expire while the project stays active; top-ups cost $0.14 per minute.

Does Keboola support CDC?

Keboola supports Debezium-based CDC for MySQL 5.7+, MariaDB 11.1.2+, and PostgreSQL 10+ through self-service connectors. PostgreSQL remains in public beta and needs wal_level=logical plus a replication slot. Additional databases are available on request:

• Oracle
• MongoDB
• SQL Server
• Cassandra

Is Keboola an ETL or ELT tool?

Keboola supports both ETL and ELT workflows, combining extraction with transformations and downstream delivery. In my Free-plan pipeline, I loaded files into Storage before running Snowflake SQL, following the ELT pattern. Free also supports Python transformations, with Flow Builder connecting extraction and transformation jobs into scheduled pipelines.