diff --git a/src/circom_adapter.cpp b/src/circom_adapter.cpp
index c2a590d..0560753 100644
--- a/src/circom_adapter.cpp
+++ b/src/circom_adapter.cpp
@@ -4,6 +4,17 @@
 #include <vector>
 #include <sstream>
 #include <stdexcept>
+#include <mutex>
+
+Circom_Circuit* getCachedCircuit(const ConstBytes& circuit_bytes) {
+    // The circuit is immutable, compiled-in data and has no destructor that
+    // frees its internal buffers, so load it once and keep it for the process
+    // lifetime instead of allocating (and leaking) it on every call.
+    static Circom_Circuit* cached = nullptr;
+    static std::once_flag once;
+    std::call_once(once, [&]() { cached = loadCircuit(circuit_bytes); });
+    return cached;
+}
 
 Circom_Circuit* loadCircuit(const ConstBytes& circuit_bytes) {
     Circom_Circuit* circuit = new Circom_Circuit;
diff --git a/src/circom_adapter.hpp b/src/circom_adapter.hpp
index e90e0c6..bd4798f 100644
--- a/src/circom_adapter.hpp
+++ b/src/circom_adapter.hpp
@@ -7,6 +7,16 @@
 
 // Return value
 Circom_Circuit* loadCircuit(const ConstBytes& circuit);
+
+// Returns a process-wide, lazily-loaded circuit for this library.
+//
+// `loadCircuit` allocates the circuit's internal buffers (input hash map,
+// witness->signal list, constants, IO field defs) but `Circom_Circuit` has no
+// destructor, so `delete circuit` leaks all of them. The circuit is immutable
+// data derived from the compiled-in `.dat`, so we load it exactly once and
+// reuse it for every (including concurrent) witness-generation call.
+Circom_Circuit* getCachedCircuit(const ConstBytes& circuit);
+
 void loadJson(Circom_CalcWit *ctx, const char* inputs_json);
 void writeBinWitness(Circom_CalcWit *ctx, Bytes* output_witness);
 
diff --git a/src/poc/ffi.cpp b/src/poc/ffi.cpp
index d5a25d4..5ec2244 100644
--- a/src/poc/ffi.cpp
+++ b/src/poc/ffi.cpp
@@ -89,26 +89,23 @@ static Status validate_witness_arguments(const WitnessInput* input, const Bytes*
 static Status generate_witness_impl(const WitnessInput* input, Bytes* output) {
     const ConstBytes& circuit_bytes = input->dat;
 
-    Circom_Circuit* circuit = loadCircuit(circuit_bytes);
+    Circom_Circuit* circuit = getCachedCircuit(circuit_bytes);
     Circom_CalcWit* ctx = new Circom_CalcWit(circuit);
 
     try {
         loadJson(ctx, input->inputs_json);
     } catch (...) {
         delete ctx;
-        delete circuit;
         throw;
     }
     if (ctx->getRemaingInputsToBeSet()!=0) {
         const std::string message = "Not all inputs have been set. Only " + std::to_string(get_main_input_signal_no()-ctx->getRemaingInputsToBeSet()) + " out of " + std::to_string(get_main_input_signal_no()) + ".";
         delete ctx;
-        delete circuit;
         return status_new(StatusCode_InvalidInput, message.c_str());
     }
 
     writeBinWitness(ctx, output);
     delete ctx;
-    delete circuit;
 
     return status_ok();
 }
diff --git a/src/pol/ffi.cpp b/src/pol/ffi.cpp
index 4a9cac2..176f3f9 100644
--- a/src/pol/ffi.cpp
+++ b/src/pol/ffi.cpp
@@ -89,26 +89,23 @@ static Status validate_witness_arguments(const WitnessInput* input, const Bytes*
 static Status generate_witness_impl(const WitnessInput* input, Bytes* output) {
     const ConstBytes& circuit_bytes = input->dat;
 
-    Circom_Circuit* circuit = loadCircuit(circuit_bytes);
+    Circom_Circuit* circuit = getCachedCircuit(circuit_bytes);
     Circom_CalcWit* ctx = new Circom_CalcWit(circuit);
 
     try {
         loadJson(ctx, input->inputs_json);
     } catch (...) {
         delete ctx;
-        delete circuit;
         throw;
     }
     if (ctx->getRemaingInputsToBeSet()!=0) {
         const std::string message = "Not all inputs have been set. Only " + std::to_string(get_main_input_signal_no()-ctx->getRemaingInputsToBeSet()) + " out of " + std::to_string(get_main_input_signal_no()) + ".";
         delete ctx;
-        delete circuit;
         return status_new(StatusCode_InvalidInput, message.c_str());
     }
 
     writeBinWitness(ctx, output);
     delete ctx;
-    delete circuit;
 
     return status_ok();
 }
diff --git a/src/poq/ffi.cpp b/src/poq/ffi.cpp
index 7715c89..44c69ab 100644
--- a/src/poq/ffi.cpp
+++ b/src/poq/ffi.cpp
@@ -89,26 +89,23 @@ static Status validate_witness_arguments(const WitnessInput* input, const Bytes*
 static Status generate_witness_impl(const WitnessInput* input, Bytes* output) {
     const ConstBytes& circuit_bytes = input->dat;
 
-    Circom_Circuit* circuit = loadCircuit(circuit_bytes);
+    Circom_Circuit* circuit = getCachedCircuit(circuit_bytes);
     Circom_CalcWit* ctx = new Circom_CalcWit(circuit);
 
     try {
         loadJson(ctx, input->inputs_json);
     } catch (...) {
         delete ctx;
-        delete circuit;
         throw;
     }
     if (ctx->getRemaingInputsToBeSet()!=0) {
         const std::string message = "Not all inputs have been set. Only " + std::to_string(get_main_input_signal_no()-ctx->getRemaingInputsToBeSet()) + " out of " + std::to_string(get_main_input_signal_no()) + ".";
         delete ctx;
-        delete circuit;
         return status_new(StatusCode_InvalidInput, message.c_str());
     }
 
     writeBinWitness(ctx, output);
     delete ctx;
-    delete circuit;
 
     return status_ok();
 }
diff --git a/src/signature/ffi.cpp b/src/signature/ffi.cpp
index 94a7990..43a792b 100644
--- a/src/signature/ffi.cpp
+++ b/src/signature/ffi.cpp
@@ -89,26 +89,23 @@ static Status validate_witness_arguments(const WitnessInput* input, const Bytes*
 static Status generate_witness_impl(const WitnessInput* input, Bytes* output) {
     const ConstBytes& circuit_bytes = input->dat;
 
-    Circom_Circuit* circuit = loadCircuit(circuit_bytes);
+    Circom_Circuit* circuit = getCachedCircuit(circuit_bytes);
     Circom_CalcWit* ctx = new Circom_CalcWit(circuit);
 
     try {
         loadJson(ctx, input->inputs_json);
     } catch (...) {
         delete ctx;
-        delete circuit;
         throw;
     }
     if (ctx->getRemaingInputsToBeSet()!=0) {
         const std::string message = "Not all inputs have been set. Only " + std::to_string(get_main_input_signal_no()-ctx->getRemaingInputsToBeSet()) + " out of " + std::to_string(get_main_input_signal_no()) + ".";
         delete ctx;
-        delete circuit;
         return status_new(StatusCode_InvalidInput, message.c_str());
     }
 
     writeBinWitness(ctx, output);
     delete ctx;
-    delete circuit;
 
     return status_ok();
 }