Airbyte
What it is
An open-source data replication (ELT) platform that has repositioned itself as “an open source data replication platform and context layer for AI agents” (docs, fetched 2026-10-07). Company note: Airbyte, Inc..
Maker, ownership and history
Founded January 2020 in San Francisco by Michel Tricot and John Lafleur. Funding: USD 26M Series A led by Benchmark, announced 2021-05-25 (Airbyte blog, CEO Michel Tricot); USD 150M Series B at a USD 1.5B valuation led by Altimeter Capital and Coatue, announced 2021-12-20, with Thrive, Salesforce Ventures, Benchmark, Accel and SV Angel (Airbyte blog). The USD 5.2M Accel-led seed (2021-03) is Wikipedia-only (secondary). Airbyte Cloud launched 2022-04-05 (Airbyte blog); Enterprise Flex 2025-09. GitHub release v2.0.0 on 2025-10-15 (Wikipedia mentions 2.0 added data activation). Independent; no acquirer found.
Editions and deployment
Core (open source, self-managed), Cloud tiers Standard, Plus, Pro, and Enterprise Flex (sovereign / hybrid / on-prem data planes). Standard and Plus bill by volume (credits); Pro and Enterprise Flex by capacity (“Data Workers”). 14-day free trial. Licence: repository uses MIT and Elastic License 2.0 (ELv2) per docs; GitHub reports “NOASSERTION” for the repo licence. Pricing: https://airbyte.com/pricing
Core architecture
Connectors run as containers (Docker/Kubernetes); low-code Connector Builder and CDK for custom sources; “over 600 sources and destinations” (docs). Full-refresh and incremental sync, CDC for databases; destinations include warehouses, lakes, and vector stores. Python SDK (PyAirbyte), HTTP API and Terraform provider.
Role in an enterprise AI rollout
Two roles: classic ELT into the lakehouse, and agent data access via “Airbyte Agents” (type-safe connectors, managed credentials, low-latency search, “Connect-Ask-Act” model, Context Store). Useful for unstructured and long-tail SaaS ingestion feeding RAG. See data-pipelines-for-ai.
AI features as of October 2026
Airbyte Agents and an MCP server are listed on the docs home page; the Airbyte Agents docs (opened 2026-10-08) show Free, Individual, Team and Custom plans and a sign-up at app.airbyte.ai, and describe skills, a Context Store, semantic search and governance, but no page labels the MCP server GA or beta (unverified). Connector Builder AI assist existed from 1.0 (2024-09, Wikipedia).
Integrations
dbt, Dagster, Airflow, Prefect (orchestrators), Snowflake, BigQuery, Databricks, Postgres, Salesforce; see dagster, apache-airflow, prefect.
Strengths and weaknesses (opinion)
Strong: open source, huge long-tail connector catalogue, self-hosting for data residency. Weak: connector quality is uneven across community connectors; self-hosting costs operations effort; licence mix (ELv2) is not OSI open source for part of the code. Compare fivetran.
Self-learning
- Docs: https://docs.airbyte.com/ (free)
- Tutorials: https://airbyte.com/tutorials (free)
- GitHub repo: https://github.com/airbytehq/airbyte
Sources (fetched 2026-10-07)
- https://docs.airbyte.com/ ; https://airbyte.com/pricing
- Second pass (2026-10-08): https://airbyte.com/blog/airbyte-raised-a-26m-series-a-to-change-how-data-is-moved ; https://airbyte.com/blog/a-150m-series-b-to-power-the-movement-of-data ; https://airbyte.com/blog/announcing-airbyte-cloud ; https://docs.airbyte.com/ai-agents/get-started.md ; https://docs.airbyte.com/llms.txt
- https://en.wikipedia.org/wiki/Airbyte (secondary)
- GitHub API for airbytehq/airbyte (releases, licence)
Open items
- Agents / MCP GA status, current connector counts per tier, Airbyte total funding after 2021, the 2021 seed round and the 2020 founding date/founders (Wikipedia-only): unverified against primary pages.