Visual pipelines with 150+ operators and full Python when you want it, orchestrated on Apache Airflow with business deadlines, checkpoint restart and ordered replay. Lineage, quality rules and access policy come with the platform, so a pipeline is governed the moment it exists, and every SAP read goes through an interface SAP keeps open.
The pages an engineer opens before recommending anything, in the order they usually come up.
Batch, incremental, change-data and streaming per connector, prerequisites, throughput reference and an SAP-compliance note nobody else shows.
Connector directory →150+ transformation, quality and load operators with inputs, outputs and examples; the Python Script Transformer and the PyLima REST server for custom code.
Documentation →A synthetic ECC MRP dataset, rows per minute per extraction method, hours to first load, run-window adherence, the 64-vCPU reference configuration; methodology first, results next.
Backbone benchmark →OpenLineage events, Open Data Contract Standard, an MCP server, Airflow 3, dbt interoperability, Iceberg and Delta output — shipped, in progress or scheduled, with dates.
Standards page →Trigger runs, read status and lineage, register sources and post exceptions from your own tooling.
Documentation →A hosted estate with SAP-shaped tables, a planning target and the incident replay, to run before any call.
Sandbox (planned) →Pipelines are Airflow DAGs; data lands as native tables in your platform and as Parquet on your storage; lineage is exported as OpenLineage events. The exit path exists on day one.
Lima's pipeline agent starts by proposing changes for you to approve; it earns the right to act on its own only through passing evaluations and a human sign-off, every write it makes is acknowledged by the target, and one click takes the right away again.
Managed Airflow inside the platform, checkpoint restart, ordered replay, deadline alerts to Slack, Teams or PagerDuty, and a run dashboard that shows margin against each window.
Table reads through a named function module, BAPIs, CDS views and change-data capture; the ODP RFC interface is absent from every Lima path.
Real-time debug on the canvas, per-task logs, lineage-linked root cause, and exception records that carry the failing row and the rule that caught it.
Drift is detected before the load and routed as an exception to the pipeline owner; the run pauses and resumes from its checkpoint.
02:14 anomaly → three pipelines auto-held → 02:51 root cause → 02:58 resumed from checkpoint → 841,000 events replayed in order → 06:30 deadline met with 3h 32m to spare.
Orchestration →A sizing estimate on 64 vCPU for a nine-source estate; the public benchmark publishes the method first and the measured results next.
Scale and volume →Moved in production on the engine, with 25M+ parallel ingress and egress loads supported.
Deployment →One engine underneath every page in this family. It reads from any source and writes to any target, runs batch loads or change-data streams, handles a nightly delta or a multi-terabyte migration, and is licensed by capacity, so the bill stays the same however many rows move. Lima's fabric answers questions on data in place; Lima Backbone moves, transforms and returns data on a deadline. When to federate and when to pipeline · The Backbone hub.
Apache Airflow. Lima tasks run inside standard DAGs, your enterprise scheduler can trigger them, and the Airflow 3 statement on the standards page says which version and when.
Yes. The Python Script Transformer runs inside a pipeline; the PyLima REST server hosts custom services; the API triggers and reads everything.
A record that fails a rule becomes an exception with the failing row, the rule, an owner and a deadline. It routes through the rules engine to a buyer, planner or steward, and is replayed after the fix.
Table reads through a named function module, BAPIs and released function modules, CDS views on S/4HANA, and change-data capture from timestamps, change pointers or triggers. RFC_READ_TABLE is barred in production and Lima's paths stay outside the ODP RFC interface.
SaaS, your private cloud or on-prem, in tiers from 32 to 128 vCPU, with SSO and row-level security in every tier.
OpenLineage events per run and per task are on the standards schedule; column-level lineage is available in the product today and through the API.
The docs and the showcase estate are open; the first load on your own data is a four-week, fixed-price engagement with a reconciliation report at the end.
