Skip to content

Move from Cognee in five minutes

HotMem reads Cognee's COGX export format. Cognee's current COGX writer emits a directory with manifest.json and JSONL files for each record kind; its optional .cogx.tar.gz archive is also accepted. See Cognee's COGX record definitions and export implementation.

1. Export your Cognee dataset

Run this against the dataset in your existing Cognee environment:

import asyncio
import cognee


async def main():
    await cognee.export(
        dataset="my_dataset",
        format="cogx",
        destination="./cognee-export",
    )


asyncio.run(main())

Do not pass include_permissions=True. Cognee's optional permissions.json contains account details and password hashes. HotMem ignores that file.

2. Import into a new HotMem database

hotmem import --from cognee \
  --db ./cognee-export \
  --target ./hotmem.sqlite \
  --out ./cognee-review.jsonl

You can pass the .cogx.tar.gz archive instead of the export directory. HotMem checks the manifest, record counts, archive paths, and each record checksum when one is present.

Review cognee-review.jsonl, then try a local query:

hotmem search "a fact from the Cognee dataset" --db ./hotmem.sqlite --json

HotMem maps typed documents, episodes, entities, facts, memories, memory blocks, and text-bearing raw graph nodes into textual memories. It preserves source IDs, scope, tags, and supported provenance in metadata. It does not rebuild Cognee's graph links as a graph. Source embeddings are not reused; to use the local semantic embedder, pass --embedder local-semantic during import after provisioning the model.

The import reads local files and does not call an LLM or hosted provider.