diff --git a/main.c b/main.c
index db17328..ded53cc 100644
--- a/main.c
+++ b/main.c
@@ -7,7 +7,7 @@
 #include "mmpriv.h"
 #include "ketopt.h"
 
-#define MM_VERSION "2.21-dev-r1084-dirty"
+#define MM_VERSION "2.21-dev-r1087-dirty"
 
 #ifdef __linux__
 #include <sys/resource.h>
@@ -73,6 +73,7 @@ static ko_longopt_t long_options[] = {
 	{ "mask-len",       ko_required_argument, 346 },
 	{ "rmq",            ko_optional_argument, 347 },
 	{ "qstrand",        ko_no_argument,       348 },
+	{ "cap-kalloc",     ko_required_argument, 349 },
 	{ "help",           ko_no_argument,       'h' },
 	{ "max-intron-len", ko_required_argument, 'G' },
 	{ "version",        ko_no_argument,       'V' },
@@ -227,6 +228,7 @@ int main(int argc, char *argv[])
 		else if (c == 345) opt.alt_drop = atof(o.arg); // --alt-drop
 		else if (c == 346) opt.mask_len = mm_parse_num(o.arg); // --mask-len
 		else if (c == 348) opt.flag |= MM_F_QSTRAND | MM_F_NO_INV; // --qstrand
+		else if (c == 349) opt.cap_kalloc = mm_parse_num(o.arg); // --cap-kalloc
 		else if (c == 330) {
 			fprintf(stderr, "[WARNING] \033[1;31m --lj-min-ratio has been deprecated.\033[0m\n");
 		} else if (c == 314) { // --frag
diff --git a/map.c b/map.c
index c34475a..1999c10 100644
--- a/map.c
+++ b/map.c
@@ -362,7 +362,9 @@ void mm_map_frag(const mm_idx_t *mi, int n_segs, const int *qlens, const char **
 		if (mm_dbg_flag & MM_DBG_PRINT_QNAME)
 			fprintf(stderr, "QM\t%s\t%d\tcap=%ld,nCore=%ld,largest=%ld\n", qname, qlen_sum, kmst.capacity, kmst.n_cores, kmst.largest);
 		assert(kmst.n_blocks == kmst.n_cores); // otherwise, there is a memory leak
-		if (kmst.largest > 1U<<28) {
+		if (kmst.largest > 1U<<28 || (opt->cap_kalloc > 0 && kmst.capacity > opt->cap_kalloc)) {
+			if (mm_dbg_flag & MM_DBG_PRINT_QNAME)
+				fprintf(stderr, "[W::%s] reset thread-local memory after read %s\n", __func__, qname);
 			km_destroy(b->km);
 			b->km = km_init();
 		}
diff --git a/minimap.h b/minimap.h
index 0b200ad..a7a886b 100644
--- a/minimap.h
+++ b/minimap.h
@@ -168,6 +168,7 @@ typedef struct {
 	int32_t max_occ, max_max_occ, occ_dist;
 	int64_t mini_batch_size; // size of a batch of query bases to process in parallel
 	int64_t max_sw_mat;
+	int64_t cap_kalloc;
 
 	const char *split_prefix;
 } mm_mapopt_t;