diff --git a/pasp/approx.c b/pasp/approx.c
index 814356f..b2748c4 100644
--- a/pasp/approx.c
+++ b/pasp/approx.c
@@ -7,7 +7,8 @@
 #include "caseo.h"
 #include "cground.h"
 
-static bool _aseo_maxent(program_t *P, size_t n_samples, size_t scale, double **R) {
+static bool _aseo_maxent(program_t *P, size_t n_samples, size_t scale, double **R, bool quiet,
+    bool status) {
   bool ok = false;
   double *a, *b, *r = a = b = NULL;
 
@@ -26,7 +27,7 @@ static bool _aseo_maxent(program_t *P, size_t n_samples, size_t scale, double **
   /* Run first ASEO separately to get the number of models. */
   models_t M = {0};
   if (!aseo_reuse(P, n_samples, MAXENT_SEMANTICS, NULL, (int) scale, 0, W, U, &M,
-        approx_rec_query_maxent)) goto cleanup;
+        approx_rec_query_maxent, false)) goto cleanup;
   size_t n_M = M.n;
 
   /* The resulting (flattened) array has dimension n*k x 1, where n is the number of queries and k
@@ -44,11 +45,18 @@ static bool _aseo_maxent(program_t *P, size_t n_samples, size_t scale, double **
   approx_query_maxent_ab(P, &M, a, b);
   approx_query_maxent_r(P, &M, r, a, b);
   models_free_contents(&M);
+  if (!quiet) {
+    for (size_t i = 0; i < P->Q_n; ++i) {
+      print_query(P->Q+i);
+      wprintf(L" = %f\n", r[i]);
+    }
+    fputws(L"---\n", stdout);
+  }
   r += n_M;
 
   for (size_t i = 1; i < m_neural; ++i) {
     if (!aseo_reuse(P, n_samples, MAXENT_SEMANTICS, NULL, (int) scale, i, W, U, &M,
-          approx_rec_query_maxent)) {
+          approx_rec_query_maxent, status)) {
       models_free_contents(&M);
       goto cleanup;
     }
@@ -57,6 +65,13 @@ static bool _aseo_maxent(program_t *P, size_t n_samples, size_t scale, double **
     approx_query_maxent_ab(P, &M, a, b);
     approx_query_maxent_r(P, &M, r, a, b);
     models_free_contents(&M);
+    if (!quiet) {
+      for (size_t i = 0; i < P->Q_n; ++i) {
+        print_query(P->Q+i);
+        wprintf(L" = %f\n", r[i]);
+      }
+      fputws(L"---\n", stdout);
+    }
     r += n_M;
   }
 
@@ -88,7 +103,7 @@ static PyObject* _aseo(PyObject *self, PyObject *args, PyObject *kwargs) {
     if (P.stable) if (!ground_all(P.stable, NULL)) goto cleanup;
   }
 
-  if (!_aseo_maxent(&P, n_samples, scale, &R)) goto cleanup;
+  if (!_aseo_maxent(&P, n_samples, scale, &R, quiet, status)) goto cleanup;
 
   bool has_neural = P.NA_n+P.NR_n > 0;
   npy_intp dims[3] = {P.Q_n, 1, 1};
diff --git a/pasp/caseo.c b/pasp/caseo.c
index 7597bef..410dac2 100644
--- a/pasp/caseo.c
+++ b/pasp/caseo.c
@@ -4,6 +4,7 @@
 #include <math.h>
 
 #include "cutils.h"
+#include "../progressbar/statusbar.h"
 
 bool watch_minimize(clingo_weight_t p, const clingo_weighted_literal_t *W, size_t n, void *data) {
   void **pack = (void**) data;
@@ -228,10 +229,11 @@ bool set_upper_bound(clingo_backend_t *back, clingo_weighted_literal_t *W, size_
 bool aseo_reuse(program_t *P, size_t k, psemantics_t psem, observations_t *O, int scale,
     size_t neural_idx, clingo_weighted_literal_t *W, clingo_weighted_literal_t *U, models_t* M,
     bool (*f)(const clingo_model_t*, program_t*, models_t*, size_t, observations_t*,
-      clingo_control_t*)) {
+      clingo_control_t*), bool status) {
   bool ok = false;
   clingo_control_t *C = NULL;
   clingo_backend_t *back = NULL;
+  statusbar *bar = status ? statusbar_new("Querying") : NULL;
 
   size_t n = num_prob_params(P);
   if (!(W && U)) {
@@ -291,6 +293,7 @@ bool aseo_reuse(program_t *P, size_t k, psemantics_t psem, observations_t *O, in
 
     if (!set_upper_bound(back, W, n, U, (int) cost)) goto cleanup;
     if (!aseo_solve(P, C, k, &res, &m, M, O, neural_idx, f)) goto cleanup;
+    if (bar) statusbar_inc(bar);
   }
   M->m = m;
 
@@ -309,19 +312,20 @@ bool aseo_reuse(program_t *P, size_t k, psemantics_t psem, observations_t *O, in
   if (clingo_error_code() != clingo_error_success) raise_clingo_error(NULL);
   else if (!ok) PyErr_SetString(PyExc_RuntimeError, "an error has occurred during ASEO!");
   clingo_control_free(C);
+  if (bar) statusbar_finish(bar);
   if (!ok) { models_free(M); M = NULL; }
   return ok;
 }
 
 bool aseo(program_t *P, size_t k, psemantics_t psem, observations_t *O, int scale,
     size_t neural_idx, models_t *M, bool (*f)(const clingo_model_t*, program_t*, models_t*, size_t,
-      observations_t*, clingo_control_t*)) {
+      observations_t*, clingo_control_t*), bool status) {
   bool ok = false;
   size_t n = num_prob_params(P);
   clingo_weighted_literal_t *W = (clingo_weighted_literal_t*) malloc(n*sizeof(clingo_weighted_literal_t));
   clingo_weighted_literal_t *U = (clingo_weighted_literal_t*) malloc(n*sizeof(clingo_weighted_literal_t));
   if (!(W && U)) goto nomem;
-  if (!aseo_reuse(P, k, psem, O, scale, neural_idx, W, U, M, f)) goto cleanup;
+  if (!aseo_reuse(P, k, psem, O, scale, neural_idx, W, U, M, f, status)) goto cleanup;
 
   ok = true;
   goto cleanup;
diff --git a/pasp/caseo.h b/pasp/caseo.h
index dadf192..e3d0c05 100644
--- a/pasp/caseo.h
+++ b/pasp/caseo.h
@@ -13,11 +13,11 @@
 
 bool aseo(program_t *P, size_t k, psemantics_t psem, observations_t *O, int scale,
     size_t neural_idx, models_t* M, bool (*f)(const clingo_model_t*, program_t*, models_t*, size_t,
-      observations_t*, clingo_control_t*));
+      observations_t*, clingo_control_t*), bool status);
 
 bool aseo_reuse(program_t *P, size_t k, psemantics_t psem, observations_t *O, int scale,
     size_t neural_idx, clingo_weighted_literal_t *W, clingo_weighted_literal_t *U, models_t *M,
     bool (*f)(const clingo_model_t*, program_t*, models_t*, size_t,
-      observations_t*, clingo_control_t*));
+      observations_t*, clingo_control_t*), bool status);
 
 #endif
diff --git a/pasp/grammar.lark b/pasp/grammar.lark
index 0e08061..e426a03 100644
--- a/pasp/grammar.lark
+++ b/pasp/grammar.lark
@@ -158,6 +158,11 @@ SEMANTICS_OPT_PROB:  "maxent" | "credal"
 _semantics_exp: ((SEMANTICS_OPT_LOGIC ("," SEMANTICS_OPT_PROB)?) | (SEMANTICS_OPT_PROB ("," SEMANTICS_OPT_LOGIC)?))
 semantics: "#semantics" (("(" _semantics_exp ")") | (_semantics_exp)) "."
 
+// Inference directive.
+exact_inf: "exact"
+aseo_inf: "aseo" "," "nmodels" "=" ID
+inference: "#inference" (exact_inf | aseo_inf) "."
+
 // Constraint.
 constraint: ":-" body "."
 
@@ -173,7 +178,7 @@ query: "#query" (("(" _interp_exp ")") | ( _interp_exp )) "."?
 // Constant definition.
 constdef: "#const" WORD "=" ID "."
 
-plp: (constdef | _fact | _rule | _ad | _neural | data | python | constraint | query | learn | semantics | _aggr)*
+plp: (constdef | _fact | _rule | _ad | _neural | data | python | constraint | query | learn | semantics | _aggr | inference)*
 
 COMMENT: "%" /[^\n]*/ NEWLINE
 
diff --git a/pasp/grammar.py b/pasp/grammar.py
index 2df07bd..9534107 100644
--- a/pasp/grammar.py
+++ b/pasp/grammar.py
@@ -474,6 +474,10 @@ def learn(self, L):
     data = self.torch_scope[L[0][1]] if L[0][0] == "PY_FUNC" else StableTransformer.path2obs(L[0][1])
     return self.pack("directive", "", ("learn", data, A))
 
+  def exact_inf(self, I): return ("inference", "exact", tuple())
+  def aseo_inf(self, I): return ("inference", "aseo", (I[0][2],))
+  def inference(self, I): return self.pack("directive", "", I[0])
+
   # Semantics directive and options.
   def SEMANTICS_OPT_LOGIC(self, _): return lark.visitors.Discard
   def SEMANTICS_OPT_PROB(self, O): return str(O)
@@ -503,7 +507,7 @@ def plp(self, C) -> Program:
     # Actual neural rules and neural ADs.
     NR, NA = [], []
     # Directives.
-    directives = {}
+    directives = {"inference": ("exact", tuple())}
     # Mapping.
     M = {"pfact": PF, "prule": PR, "query": Q, "varquery": VQ, "cfact": CF, "ad": AD, "nrule": TNR,
          "nad": TNA}
diff --git a/pasp/program.py b/pasp/program.py
index fca1646..e1a8fb2 100644
--- a/pasp/program.py
+++ b/pasp/program.py
@@ -426,7 +426,11 @@ def __call__(self, **kwargs):
         else: learn(self, D, **A)
     if len(self.Q) + len(self.VQ) > 0:
       from exact import exact
+      from approx import aseo
       A = {"quiet": False, "status": True}
       A.update(kwargs)
-      if "psemantics" in self.directives: A.update(self.directives["psemantics"])
-      return exact(self, **A)
+      # TODO: implement additional semantics for ASEO and remove the exact exception below.
+      if ("psemantics" in self.directives) and (self.directives["inference"][0] == "exact"):
+        A.update(self.directives["psemantics"])
+      f = vars()[self.directives["inference"][0]]
+      return f(self, *self.directives["inference"][1], **A)