Overview
What is Kiwi Fabric?
What Kiwi Fabric does, the problems it solves, and its main capabilities.
Kiwi Fabric is a self-hosted engine that presents many data systems as one data surface. You write standard SQL against tables that may physically live in PostgreSQL, SQL Server, MongoDB, a REST API or other registered sources. Kiwi Fabric resolves each table to the right source, fetches only what is needed, and combines the results. It also coordinates multi-system writes as a single transaction-like unit.
Data is never copied into Kiwi Fabric. Every result is fetched live from the originating system at query time.
The problem it addresses#
Most organisations keep data in several systems that were never designed to work together. That makes two things hard:
- Reading across systems. The usual answers are ETL pipelines into a warehouse (always somewhat stale, and a team to maintain them), application code that fans out requests and joins results by hand, or giving every BI tool direct access to every system.
- Writing across systems. A single business operation, such as "place an order, deduct inventory, create a shipping record", may touch several databases. Two-phase commit rarely works across different vendors, and full event sourcing is a large investment. Many teams fall back on manual reconciliation.
Every consuming tool also wants a different protocol: BI tools want the PostgreSQL wire protocol, data-science tools want Arrow Flight, and applications want REST/JSON.
What you get#
Federated SQL reads. Joins, filters, grouping, sorting and paging across sources in one statement. See Run federated queries.
Distributed writes with compensation. A saga is a declarative JSON sequence of inserts, updates and deletes across registered sources. It either completes, or every completed step is undone in reverse order. See Run a distributed transaction.
Three network surfaces from one engine. REST/JSON, PostgreSQL wire protocol and Arrow Flight SQL, all serving the same live data and the same security rules. See Network surfaces.
Row- and column-level security. Role-based access control, plus row filters and column masks, enforced centrally on every surface. See Secure data access.
AI-assisted access. Register one or more LLM providers and turn natural-language questions into reviewable SQL, or describe a transaction and get a saga draft. See Generate SQL and sagas with AI.
Visibility into gaps. Not every system can be rolled back automatically. Kiwi Fabric reports exactly which write paths have no compensation. See Compensation gaps.
Observability and governance. Every query and saga is recorded with per-source timings, and lineage shows dependencies between virtual tables and sources. See Observability and telemetry.
How it is deployed#
Kiwi Fabric runs as a single self-hosted process per customer, a silo model. There is no shared execution environment or connection pool between customers. Handler packages are loaded on demand, so you only deploy the data-source support you need. Cloud and on-premises deployment are both supported.
Management interfaces#
You can manage Kiwi Fabric through the Web UI (data sources, capabilities, virtual tables, saga templates, the saga runner, the SQL playground, security, resource guards, AI assistants, lineage, telemetry), the kiwi fabric command line (CLI reference), the REST API (REST API reference), and the client SDKs for Python, .NET, Java and Node.js.
A tour of the Web UI#
The left rail groups the console into Execute (Playground and Saga Runner), Assets, Metadata, Monitor, Administration and Settings. Each group opens a menu of pages, and the buttons across the top bar are quick links to the pages in the group you are working in.
Use the search box in the top bar to find a page, and the ? button for help written for the page you are on.
