Airbyte
Also known as: Airbyte Cloud, Airbyte Context Layer, Airbyte MCP
Airbyte moves data from operational systems into warehouses and gives agents a way to use that data. The hosted context layer supports live reads from configured sources and search across selected data that Airbyte has already synced. Over MCP, an assistant can find relevant records, inspect a failed sync and manage Airbyte connections within its assigned permissions. Airbyte also offers customer run replication and hybrid deployments for teams that keep movement infrastructure inside their own environment.
Airbyte supplies a data replication platform and a context layer for teams that build agents on operational data. A connection moves data from a source to a destination, such as a database table or a data warehouse. The catalog includes more than 700 connectors for systems such as Postgres, Salesforce, Zendesk and Snowflake. A team can build additional connectors and schedule syncs on an interval or an exact calendar cadence.
Incremental connections can send newly added or updated rows, while a full refresh sends the entire selected dataset. More than 25,000 companies use Airbyte, including one in five of the Fortune 500, and it moves 26 billion records across 2 million pipeline syncs a day. The open source project has more than 20,000 GitHub stars.
Airbyte Cloud supports two ways for agents to find data through MCP. Agent access reads from supported systems using the connector's stored credentials. Search builds an index from data synced through a selected connection and matches both exact terms and meaning. Administrators can enable either capability on its own. For a support analysis, an assistant might search ticket text for recurring setup issues and then query the warehouse to group the affected accounts. Backfill can add previously synced content for eligible Snowflake and BigQuery destinations. The same MCP connection can manage Airbyte pipelines, start a sync and retrieve its logs.
Airbyte Core is free for customer run replication under the Elastic License 2.0. The hosted Standard plan starts at $20 per month with five credits and syncs as often as hourly. Plus starts at $189 per month for 40 credits, rising to 2,000 credits at $4,999, with syncs every 15 minutes, single sign on and up to two workspaces. Pro and Enterprise Flex use quoted compute capacity measured in Data Workers, with syncs every five minutes and role based access.
Airbyte holds SOC 2 Type II attestation and ISO/IEC 27001:2022 certification. Agent access is off until an organization administrator enables it. Hosted workspaces can run in the United States or European Union, and Enterprise Flex can place a data plane inside the buyer's environment, on premises or across clouds. Hosted context search and direct reads are unavailable on customer owned data planes.
Airbyte fits teams that need to keep warehouses current and give an existing assistant access to selected business data. Customers include Wise, Kuda Bank, Air Liquide, monday.com, Miro, Unity and Peloton. Perplexity uses Airbyte with PostgreSQL and Snowflake to manage product data. WeWork moved data from Salesforce, databases and internal applications into Snowflake and cut projected data movement costs by two thirds.
Teams using the agent tools can search synced support records, compare those results with structured tables or inspect the connection when a scheduled refresh fails. A workspace administrator determines which data is accessible, and the agent operates inside the limits of that configuration. Direct reads retrieve data from the connected system, while the source and destination remain the systems of record.
Vendor details
Canonical URL
https://airbyte.com
Category
Agent infrastructure
Subcategory
Data replication and managed business context for agents
Funding status
Airbyte, Inc. is a privately held software company.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
Airbyte replication connects databases, warehouses and SaaS sources. The Cloud context layer supports live access and search through Airbyte MCP with clients including Claude, ChatGPT and VS Code Copilot.
In practice
A support analyst uses an assistant to find similar Zendesk tickets in an Airbyte search index and query Snowflake to group the affected accounts by plan.
A data engineer has an assistant inspect the logs for a failed Salesforce sync and then start another sync after a connection setting is corrected.
A company with a customer owned data plane runs its replication with Enterprise Flex and reserves agent search for workspaces on Airbyte managed infrastructure.
Sources & related URLs
Related / legacy domains
Research sources
Agentic Index coverage score
7.5 / 14 capabilities · 54%
| Integrations & Tool Calling | Partial |
|---|---|
|
Agents can use configured third party connectors to retrieve records from operational systems and query supported warehouses. An agent also can start an Airbyte sync that moves records to a destination. Direct calls into the source or destination through the context tools are read only. The replication connection determines what Airbyte writes during a sync. Sourcedocs.airbyte.com/platform/airbyte-mcpread 2026-11-20 |
|
| Workflow Orchestration | Partial |
|
Replication runs as a managed connection that reads selected streams and writes the resulting records to a destination. A calling assistant can create the connection, select streams and start the process through Airbyte tools. Airbyte executes the configured movement of data. The assistant or its workflow system decides how to sequence that operation with any further actions. Sourcedocs.airbyte.com/platform/airbyte-mcpread 2026-11-20 |
|
| Knowledge Grounding & RAG | Full |
|
Airbyte builds a searchable index for selected data as a connection syncs it. A query can match exact terms or meaning in records such as tickets and conversation text. It returns the matched text and source record so an assistant can join the information to a warehouse table. For eligible Snowflake or BigQuery connections, a backfill can index data that was synced previously. Sourcedocs.airbyte.com/platform/context-layer/searchread 2026-11-20 |
|
| Human Oversight & Guardrails | Partial |
|
Airbyte protects destructive pipeline management actions by requiring a human to mark a resource before the agent can delete it. The local demonstration mode can further limit changes to objects created in the current session. These safeguards constrain some of the tools the assistant can run. The outside client owns its own approval prompts during an agent conversation. Sourcedocs.airbyte.com/platform/airbyte-mcp/toolsread 2026-11-20 |
|
| Security, Identity & Governance | Full |
|
Airbyte holds SOC 2 Type II attestation and ISO/IEC 27001:2022 certification. Organization administrators decide whether agents can access data. Workspace editors can further scope the sources and destinations. Higher plans support single sign on, SCIM and roles. A tool checks the Airbyte user's permissions, while a live read uses the permissions of the connector's stored credentials. Sourceairbyte.com/securityread 2026-11-20 |
|
| Observability & Auditability | Partial |
|
A workspace can show the current status of its syncs, and a developer can see recent jobs, their results and the volume moved through a particular connection. A tool can also retrieve the log from an individual job when something fails. The job log covers the pipeline execution. The connected assistant's reasoning and its other actions belong to the client that initiated the request. Sourcedocs.airbyte.com/platform/airbyte-mcp/toolsread 2026-11-20 |
|
| Memory & State Persistence | Not documented |
|
The context layer reads source records on demand or retrieves data from a connection's search index. Synced business records remain available for later requests, and an administrator controls the connections that are indexed. The assistant keeps its own conversation history. Airbyte stores business records for retrieval rather than saving an agent's learned task state across runs. Sourcedocs.airbyte.com/platform/context-layerread 2026-11-20 |
|
| Deployment & Data Residency | Full |
|
A hosted workspace runs its connections in a selected United States or European Union region. Enterprise Flex can place the replication data plane in the buyer's environment while sending metadata to Airbyte's hosted control plane. Workspaces with customer owned data planes cannot enable the hosted direct reads or search that the Cloud context layer provides. Sourceairbyte.com/securityread 2026-11-20 |
|
| Prebuilt Agents / Templates / Packs | Partial |
|
The catalog offers prepared connectors for databases, warehouses and SaaS tools. A direct read tool can also retrieve a connector skill that describes its entities, supported actions and parameters before using it. A developer chooses and authorizes the connectors that form the agent's data access. The assistant supplies the task and uses those connector tools within its own workflow. Sourcedocs.airbyte.com/platform/airbyte-mcp/toolsread 2026-11-20 |
|
| Triggers & Channel Coverage | Partial |
|
A connection can refresh on a calendar interval or a schedule defined by a cron expression. When eligible data syncs, the managed search index can reflect the new records. A user or a caller can also start a sync manually. The schedule runs the replication task. Agent queries start when an outside client invokes a tool with the particular source or pipeline request. Sourcedocs.airbyte.com/platform/using-airbyte/core-concepts/sync-schedulesread 2026-11-20 |
|
| Model Flexibility & Routing | Not documented |
|
The hosted search index embeds the question and the synced content and ranks matching passages. An assistant chooses the search, direct read or query tool based on its task. The assistant's model belongs to the calling client, and Airbyte manages the embeddings and reranking inside search. Workspace controls select the data and retrieval method used for a request. Sourceairbyte.com/blog/search-in-airbyte-context-layerread 2026-11-20 |
|
| APIs / SDKs / MCP Extensibility | Full |
|
The public API addresses the sources, destinations and connections of an Airbyte deployment. An assistant can also operate those resources through the Airbyte MCP server, including creating a connection or invoking an existing sync. Cloud and supported self managed deployments expose programmatic management of their own resources, with access tied to registered applications and user roles. Sourcedocs.airbyte.com/developers/api-documentationread 2026-11-20 |
|
| Testing, Debugging & Optimization | Partial |
|
A developer can inspect individual connection runs, see which sync failed and use the job logs to diagnose the disagreement between intended configuration and what the pipeline did. The connector interface returns job status and recent history for comparison. These tools let the team debug data flow and connectivity. Evaluating the outputs of an agent that consumes those records happens outside Airbyte. Sourcedocs.airbyte.com/platform/airbyte-mcp/toolsread 2026-11-20 |
|
| Browser / Computer-use | Not documented |
|
The context layer uses authenticated data connectors to read live records and a search index to return passages from synchronized data. An assistant works with those records through tools and receives structured output. Requests use the existing source or warehouse integration. The context layer does not drive the application interface or operate a browser to retrieve those records. Sourcedocs.airbyte.com/platform/context-layerread 2026-11-20 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Pricing
Core free to run yourself, Cloud Standard from $20 a month
Monthly credits by volume on Standard and Plus, Data Workers on Pro and Enterprise Flex
What is public
Airbyte charges from $20 a month for Cloud Standard with five credits, about 1 GB of syncs, and hourly syncs. Plus starts at $189 a month for 40 credits and steps through 100 credits at $449, 250 at $999, 500 at $1,799, 1,000 at $3,199 and 2,000 at $4,999, with syncs every 15 minutes and single sign on. Airbyte Core is free to run yourself. Hosted trials end at 14 days or 400 credits, about 100 GB of syncs, whichever comes first.
Billing mechanics
Standard and Plus credit usage depends on the source. API and custom sources use six credits per million rows. Database and file sources use four credits per GB processed. Pro and Enterprise Flex charge for capacity measured in Data Workers.
Variable cost rationale
Standard and Plus bill by rows or gigabytes synced, and usage past the plan's credits is billed monthly while schedules keep running, so a full refresh or a busy source can raise the bill quickly.
Additional watchouts
After the plan credits run out, scheduled syncs continue and additional usage is billed monthly. A full refresh counts every row that it syncs.
Sales call required
Mixed (some tiers require a call)
Free / trial
Airbyte Core free, Cloud trial for 14 days or 400 credits
Lowest paid plan
Cloud Standard
Missing data
Airbyte quotes Pro and Enterprise Flex capacity rates.
Related vendors
- AgentOps — Agent observability and debugging platform: open source SDKs trace…
- AgentPhone — AgentPhone gives the AI agents a team builds their own US or…
- Agno — Python agent framework and AgentOS runtime (formerly Phidata) for…
- AIsa — Resource and payment gateway for AI agents, with one key to 110+…
- Albato — Albato is a no code integration platform that sells an embedded,…
- AlphaBitCore — AI control plane for regulated financial firms: one gateway enforces…
Alternatives to Airbyte
Other products that do the work Airbyte does, and what each one is.
- Reducto Office
- Bird
- TeradataData layer agents run
- AssemblyAI
- CorpusIQRead only business data layer for AI assistants
- GentoroMCP platform