diff --git a/pkg/index/corpus.go b/pkg/index/corpus.go
new file mode 100644
index 000000000..a147482fa
--- /dev/null
+++ b/pkg/index/corpus.go
@@ -0,0 +1,44 @@
+package index
+
+import (
+	"sync"
+
+	"camlistore.org/pkg/blob"
+)
+
+// Corpus is an in-memory summary of all of a user's blobs' metadata.
+type Corpus struct {
+	mu    sync.RWMutex
+	strs  map[string]string // interned strings
+	blobs map[blob.Ref]BlobMeta
+	// TODO: add GoLLRB to third_party; keep sorted BlobMeta
+	files      map[blob.Ref]*FileMeta
+	permanodes map[blob.Ref]*PermanodeMeta
+	deletedBy  map[blob.Ref]blob.Ref // key is deleted by value
+}
+
+type BlobMeta struct {
+	Size      int
+	CamliType string
+}
+
+type FileMeta struct {
+	size      int64
+	mimeType  string
+	wholeRefs []blob.Ref
+}
+
+type PermanodeMeta struct {
+	OwnerKeyId string
+	// Claims     ClaimList
+}
+
+func NewCorpusFromStorage(s Storage) (*Corpus, error) {
+	panic("TODO")
+}
+
+func (x *Index) KeepInMemory() (*Corpus, error) {
+	var err error
+	x.corpus, err = NewCorpusFromStorage(x.s)
+	return x.corpus, err
+}
diff --git a/pkg/index/index.go b/pkg/index/index.go
index d5319d92e..d819a07b1 100644
--- a/pkg/index/index.go
+++ b/pkg/index/index.go
@@ -150,6 +150,8 @@ type Index struct {
 	// of the blobs in the index. It makes for faster reads than the otherwise
 	// recursive calls on the index.
 	deletes *deletionCache
+
+	corpus *Corpus // or nil, if not being kept in memory
 }
 
 var _ blobserver.Storage = (*Index)(nil)