diff --git a/spec/get.md b/spec/get.md
new file mode 100644
index 0000000..36d26b8
--- /dev/null
+++ b/spec/get.md
@@ -0,0 +1,43 @@
+# get / `dvs get` / `dvs::dvs_get`
+
+## CLI
+
+The option to return `--json` must be present.
+
+```shell
+$ dvs get <PATHS> [OPTIONS] --help
+
+Paths:
+    Optional list of files or directories to retrieve from storage. If missing, then all files
+    are retrieved, analogous to `dvs sync`
+
+Options:
+    --json Outputs are emitted in a JSON format
+    --glob Specify a glob expression for which tracked datafiles and/or directories to retrieve
+    --help, -h
+```
+
+`dvs get` shall accept directories as arguments, unlike `dvs add`, the activity of syncing a complete directory is within expected user workflow, while adding directories can be done unintentionally, and have ramifications on the coherence of the dvs backend/storage.
+
+## R
+
+Signature:
+
+```r
+#' @param paths,glob provide a list of files to sync from backend, either explicitly `files` or via glob expression `glob`.
+#' @param progress provide a progress bar
+#' @param parallel control number of threads to use, `FALSE`/`0`, while `TRUE` implies at most 16 threads, or provide non-zero integer for more.
+dvs_get <- function(
+  paths = character(),
+  glob = character(),
+  progress = FALSE, 
+  parallel = TRUE
+)
+```
+
+## Initial release concerns
+
+### Timeout / cancellation
+
+In case of retrieving large amount of data (either a lot of files, or big files, or both) the user may cancel the operation via `Ctrl+C`, or a connection can timeout, or similar. The `get` command ought to be able to continue from elapsed progress, and it should tell the user, that indeed,
+it is resuming a previous retrieval.