dlt Destination

dlt is the open-source Python ELT library (Apache-2.0). dlt-longmem is a destination: anything dlt can read can land in agent memory.

Use#

import dlt
from dlt_longmem import longmem

pipeline = dlt.pipeline(pipeline_name="kb",
                        destination=longmem(collection="docs"))
pipeline.run(my_source())        # rows → RememberOS, embedded + searchable

Install from the repo: pip install "git+https://github.com/11data/longmem#subdirectory=connectors/dlt-longmem". Key via LONGMEM_API_KEY or dlt secrets.

How records map#

• The text column becomes the memory body — text_field=… to choose, else the first of content/text/body/summary/description/title, else the whole row as JSON.
• Remaining columns are kept as queryable metadata.
• container_tag=… scopes every record (e.g. per customer).
• Writes batch through the bulk API, 100 at a time.

Caveat#

Memory text caps at 10,000 chars — long document bodies get truncated. For large documents use the file-based path (drop / SharePoint importer), which chunks properly instead of truncating.