Skip to content

Airbyte

Data-movement platform with 600-plus connectors that syncs SaaS apps and databases into warehouses and vector stores, self-hosted or cloud.

Reviewed by Sheela · Updated

View on GitHub Open source

Our rating

3.9/5

Airbyte has the widest open catalogue for pulling business data into a RAG index, but its AI destinations lag its ELT core and self-hosting it means owning its security patches.

What it is

Airbyte moves data from APIs, databases and files into warehouses, lakes and AI applications. Its catalogue lists more than 600 connectors, with a no-code Connector Builder and a low-code kit for writing new ones, and syncs are incremental so you do not re-import everything each time.

For retrieval pipelines, vector destinations such as Pinecone split records into chunks of up to 8,191 tokens, embed them with OpenAI, Cohere, Azure OpenAI or any OpenAI-compatible endpoint, and write them to the index. A separate Airbyte Agents product exposes connectors to agents through a hosted MCP server, SDK and CLI.

The repository carries both the MIT and Elastic License 2.0 (ELv2) licences; ELv2 forbids offering the software to others as a managed service. Self-managed Airbyte reached version 2.4 on 9 October 2026. A critical remote code execution flaw in the Connector Builder was fixed in 2.2, and 1.x will not be patched.

Our review

Airbyte's case is breadth with incremental sync. When the knowledge you need sits in Salesforce, Zendesk, Notion or Postgres rather than on the web, more than 600 connectors and scheduled incremental updates beat writing and maintaining your own API clients. The repository has about 22,300 GitHub stars and daily commits, and self-managed releases have kept a steady pace, from 2.1 in April to 2.4 on 9 October 2026, even though GitHub's release page stopped at 2.0.

The RAG layer is the weak part. The vector destinations do their own chunking and embedding, which is convenient, but the Pinecone connector's docs, updated as recently as 13 October 2026, still name text-embedding-ada-002 and Cohere's embed-english-light-v2.0 as the model options, both several generations old. The OpenAI-compatible setting lets you point at a current model; syncing to a warehouse and embedding downstream gives more control.

Security is now the main self-hosting question. On 17 October 2026 Airbyte published a critical advisory: before 2.2, any user with Workspace Reader access could run commands in the Manifest Server container through Connector Builder manifests, and on 1.x installs from the legacy Helm chart the service skipped authentication entirely. Cloud customers needed to do nothing; 1.x will get no fix. Telemetry from the UI and servers is also on unless you change the tracking strategy in values.yaml.

Operations are the other cost. Self-managed Airbyte is a multi-service deployment you upgrade and monitor yourself, and Airbyte's own docs say Marketplace connectors, maintained by the community, carry no support SLA and may take breaking changes with no notice. Cloud pricing removes that work but moves the bill from servers to credits, roughly 4 to 5 credits per GB synced on Airbyte's own examples, which at 250 GB a month is the $3,199 Plus package.

Where it earns its place

  • A data engineer at a 300-person SaaS company building an internal support copilot

    Keep Zendesk tickets, Confluence pages and Postgres records synced into a vector index

    Self-host Airbyte 2.4 with abctl on a 4-CPU, 8 GB machine, sync incrementally to a warehouse and embed downstream with a current model, or use the Pinecone destination with an OpenAI-compatible embedding endpoint. Software is free; the cost is the server, patching and connector babysitting.

  • A head of data at a mid-size retailer without platform engineers

    Avoid running sync infrastructure at all

    Airbyte Cloud Plus starts at $189 a month for 40 credits; at roughly 4 credits per GB, 250 GB a month means the 1,000-credit package at $3,199. SSO starts on Plus; data residency and row filtering need Pro, which is priced by capacity on request.

  • A security-conscious platform team at a regulated insurer

    Run connectors inside its own network with vendor support

    Enterprise Flex runs data planes in your VPC or on-premises with Airbyte's control plane; pricing is custom. Upgrade self-managed installs to 2.2 or later and restrict Connector Builder access, per the October 2026 advisory.

Before you commit

Setup
curl -LsfS https://get.airbyte.com | bash - installs abctl, then abctl local install, which can take up to 30 minutes; needs Docker and 4 or more CPUs with 8 GB RAM, or 2 CPUs in low-resource mode without the Connector Builder.
Integrations
More than 600 sources and destinations including Snowflake, BigQuery, Postgres, Pinecone and other vector stores; embedding via OpenAI, Cohere, Azure OpenAI or OpenAI-compatible endpoints; Terraform and API; a hosted MCP server at mcp.airbyte.ai for Claude, ChatGPT, Cursor and other clients.
Data and privacy
Self-hosted telemetry from the UI and servers is on unless global.tracking.strategy is set to logging. Advisory GHSA-g35w-6r6v-h4r7 (critical) allowed command execution through the Connector Builder before 2.2, and CVE-2024-38363 was an earlier template-injection RCE. Enterprise Flex keeps data in your own environment; ELv2 forbids resale as a managed service.
Cost at scale
Core is free on your own servers. Cloud: Standard from $20 a month with 5 credits (about 1 GB); Plus credit packs from 40 for $189 to 2,000 for $4,999 a month; Pro is billed by data workers, each running about three concurrent syncs, at custom prices.
Limits
About 2,660 open issues and pull requests; community connector quality varies; Standard syncs at most hourly, Plus every 15 minutes, Pro every 5; 1.x receives no security fixes.

What's good

  • More than 600 connectors with incremental sync
  • Self-hostable, with a free core
  • Vector destinations chunk and embed for you
  • Hosted MCP server for agent access through Airbyte Agents

What's not

  • Critical 2026 RCE in Connector Builder; 1.x left unpatched
  • Embedding model options in docs are outdated
  • Self-hosting is operationally heavy
  • ELv2 bars resale as a hosted service

Best for

Data teams feeding a RAG system from many SaaS apps and databases who want incremental syncs they can self-host.

Alternatives

Airbyte: common questions

Is Airbyte worth it?
Airbyte has the widest open catalogue for pulling business data into a RAG index, but its AI destinations lag its ELT core and self-hosting it means owning its security patches. We rate it 3.9 out of 5.
How much does Airbyte cost?
Self-hosted Core is free (MIT and ELv2). Cloud Standard from $20/month with 5 credits; Plus from $189/month for 40 credits up to $4,999 for 2,000; Pro and Enterprise Flex custom.
What are the downsides of Airbyte?
Critical 2026 RCE in Connector Builder; 1.x left unpatched; embedding model options in docs are outdated; self-hosting is operationally heavy; eLv2 bars resale as a hosted service.
Who is Airbyte best for?
Data teams feeding a RAG system from many SaaS apps and databases who want incremental syncs they can self-host.
How much does Airbyte cost for a team or in production?
Core is free on your own servers. Cloud: Standard from $20 a month with 5 credits (about 1 GB); Plus credit packs from 40 for $189 to 2,000 for $4,999 a month; Pro is billed by data workers, each running about three concurrent syncs, at custom prices.
Is Airbyte safe for company data?
Self-hosted telemetry from the UI and servers is on unless global.tracking.strategy is set to logging. Advisory GHSA-g35w-6r6v-h4r7 (critical) allowed command execution through the Connector Builder before 2.2, and CVE-2024-38363 was an earlier template-injection RCE. Enterprise Flex keeps data in your own environment; ELv2 forbids resale as a managed service.
What are the best Airbyte alternatives?
The closest alternatives we have reviewed are Unstructured and LlamaIndex.

Answers checked Oct 23, 2026.