diff --git a/liblfi/include/lfi.h b/liblfi/include/lfi.h
index 51a30be..c08ba3b 100644
--- a/liblfi/include/lfi.h
+++ b/liblfi/include/lfi.h
@@ -95,4 +95,6 @@ struct LFILoadOpts {
 bool                    lfi_proc_loadelf(struct LFIAddrSpace* as, uint8_t* prog, size_t progsz, uint8_t* interp, size_t interpsz, struct LFILoadInfo* o_info, struct LFILoadOpts opts);
 bool                    lfi_proc_init(struct LFIContext* ctx, struct LFIAddrSpace* as, struct LFILoadInfo info);
 
+void lfi_thread_init(void* thread_create, void* pausefn);
+
 char* lfi_strerror(void);
diff --git a/liblfi/pal/arch/arm64/regs.c b/liblfi/pal/arch/arm64/regs.c
index 5062236..2263f3a 100644
--- a/liblfi/pal/arch/arm64/regs.c
+++ b/liblfi/pal/arch/arm64/regs.c
@@ -11,3 +11,15 @@ lfi_regs_init(struct TuxRegs* regs, struct LFIAddrSpace* as, struct LFIContext*
     regs->x30 = as->base;
     regs->REG_SYS = (uintptr_t) ctx->sys;
 }
+
+uintptr_t*
+lfi_regs_entry(struct TuxRegs* regs)
+{
+    return &regs->x30;
+}
+
+uintptr_t*
+lfi_regs_arg0(struct TuxRegs* regs)
+{
+    return &regs->x0;
+}
diff --git a/liblfi/pal/arch/x64/regs.c b/liblfi/pal/arch/x64/regs.c
index 4855320..b497b60 100644
--- a/liblfi/pal/arch/x64/regs.c
+++ b/liblfi/pal/arch/x64/regs.c
@@ -9,3 +9,15 @@ lfi_regs_init(struct TuxRegs* regs, struct LFIAddrSpace* as, struct LFIContext*
     regs->rsp = as->base;
     regs->r13 = (uintptr_t) ctx->sys;
 }
+
+uintptr_t*
+lfi_regs_entry(struct TuxRegs* regs)
+{
+    return &regs->r11;
+}
+
+uintptr_t*
+lfi_regs_arg0(struct TuxRegs* regs)
+{
+    return &regs->rdi;
+}
diff --git a/liblfi/pal/ctx.c b/liblfi/pal/ctx.c
index 0b9fa6c..2faef6f 100644
--- a/liblfi/pal/ctx.c
+++ b/liblfi/pal/ctx.c
@@ -7,8 +7,11 @@
 #include "pal/platform.h"
 #include "pal/regs.h"
 
+// context for injecting clone calls
 pthread_mutex_t lfi_clonectx_lk = PTHREAD_MUTEX_INITIALIZER;
 struct LFIContext* lfi_clonectx;
+// for newly cloned thread contexts
+_Thread_local struct LFIContext* lfi_newctx;
 
 _Thread_local struct LFIContext* lfi_myctx;
 
@@ -154,8 +157,14 @@ pal_register_clonectx(struct LFIContext* ctx)
     lfi_clonectx = ctx;
 }
 
-void
-pal_register_myctx(struct LFIContext* ctx)
+EXPORT void
+lfi_thread_init(void* thread_create, void* pausefn)
 {
-    assert(!"unimplemented");
+    // invoke sbx_thread_create(&_lfi_pause) with the clone context
+    LOCK_WITH_DEFER(&lfi_clonectx_lk, lk);
+    *lfi_regs_entry(&lfi_clonectx->regs) = (uintptr_t) thread_create;
+    *lfi_regs_arg0(&lfi_clonectx->regs) = (uintptr_t) pausefn;
+    lfi_ctx_run(lfi_clonectx, lfi_clonectx->as);
+    lfi_myctx = lfi_newctx;
+    lfi_newctx = NULL;
 }
diff --git a/liblfi/pal/platform.h b/liblfi/pal/platform.h
index 2974c52..1e355e9 100644
--- a/liblfi/pal/platform.h
+++ b/liblfi/pal/platform.h
@@ -54,6 +54,8 @@ gb(size_t x)
 
 _Thread_local extern struct LFIContext* lfi_myctx;
 
-void pal_register_clonectx(struct LFIContext* ctx);
+_Thread_local extern struct LFIContext* lfi_newctx;
+
+extern struct LFIContext* lfi_clonectx;
 
-void pal_register_myctx(struct LFIContext* ctx);
+void pal_register_clonectx(struct LFIContext* ctx);
diff --git a/liblfi/pal/regs.h b/liblfi/pal/regs.h
index 76531a6..8fb51ba 100644
--- a/liblfi/pal/regs.h
+++ b/liblfi/pal/regs.h
@@ -1,3 +1,7 @@
 #include "lfi_arch.h"
 
 void lfi_regs_init(struct TuxRegs* regs, struct LFIAddrSpace* as, struct LFIContext* ctx);
+
+uintptr_t* lfi_regs_entry(struct TuxRegs* regs);
+
+uintptr_t* lfi_regs_arg0(struct TuxRegs* regs);
diff --git a/liblfi/syscalls/sys_proc.c b/liblfi/syscalls/sys_proc.c
index 0e5c0fb..0adc396 100644
--- a/liblfi/syscalls/sys_proc.c
+++ b/liblfi/syscalls/sys_proc.c
@@ -21,10 +21,11 @@ uintptr_t
 sys_exit(struct TuxThread* p, uint64_t code)
 {
     clearctid(p);
-    if (p->proc->tux->opts.pause_on_exit)
+    if (p->proc->tux->opts.pause_on_exit) {
         lfi_ctx_pause(p->p_ctx, code);
-    else
+    } else {
         lfi_ctx_exit(p->p_ctx, code);
+    }
     assert(!"unreachable");
 }
 
diff --git a/liblfi/syscalls/sys_thread.c b/liblfi/syscalls/sys_thread.c
index 4766a69..d38d7ad 100644
--- a/liblfi/syscalls/sys_thread.c
+++ b/liblfi/syscalls/sys_thread.c
@@ -132,6 +132,11 @@ spawn(struct TuxThread* p, uint64_t flags, uint64_t stack, uint64_t ptidp, uint6
         // was created so it can be reused when we need to spawn threads in the
         // future.
         pal_register_clonectx(p2->p_ctx);
+    } else if (p->p_ctx == lfi_clonectx) {
+        threadspawn(p2);
+        lfi_newctx = p2->p_ctx;
+        // does not return
+        lfi_ctx_pause(p->p_ctx, 0);
     } else {
         // Actually create a new thread.
         pthread_t thread;