cactus-react-native 0.1.0 → 0.1.2

package/ios/Cactus.mm

@@ -643,4 +643,84 @@ RCT_EXPORT_METHOD(releaseVocoder:(double)contextId
 }
 #endif
 
+// New conversation management methods
+RCT_EXPORT_METHOD(generateResponse:(double)contextId
+                 userMessage:(NSString *)userMessage
+                 maxTokens:(double)maxTokens
+                 withResolver:(RCTPromiseResolveBlock)resolve
+                 withRejecter:(RCTPromiseRejectBlock)reject)
+{
+    CactusContext *context = llamaContexts[[NSNumber numberWithDouble:contextId]];
+    if (context == nil) {
+        reject(@"llama_error", @"Context not found", nil);
+        return;
+    }
+    if ([context isPredicting]) {
+        reject(@"llama_error", @"Context is busy", nil);
+        return;
+    }
+    dispatch_async(llamaDQueue, ^{
+        @try {
+            @autoreleasepool {
+                NSString *result = [context generateResponse:userMessage maxTokens:(int)maxTokens];
+                resolve(result);
+            }
+        } @catch (NSException *exception) {
+            reject(@"llama_cpp_error", exception.reason, nil);
+        }
+    });
+}
+
+RCT_EXPORT_METHOD(continueConversation:(double)contextId
+                 userMessage:(NSString *)userMessage
+                 maxTokens:(double)maxTokens
+                 withResolver:(RCTPromiseResolveBlock)resolve
+                 withRejecter:(RCTPromiseRejectBlock)reject)
+{
+    CactusContext *context = llamaContexts[[NSNumber numberWithDouble:contextId]];
+    if (context == nil) {
+        reject(@"llama_error", @"Context not found", nil);
+        return;
+    }
+    if ([context isPredicting]) {
+        reject(@"llama_error", @"Context is busy", nil);
+        return;
+    }
+    dispatch_async(llamaDQueue, ^{
+        @try {
+            @autoreleasepool {
+                NSDictionary *result = [context continueConversation:userMessage maxTokens:(int)maxTokens];
+                resolve(result);
+            }
+        } @catch (NSException *exception) {
+            reject(@"llama_cpp_error", exception.reason, nil);
+        }
+    });
+}
+
+RCT_EXPORT_METHOD(clearConversation:(double)contextId
+                 withResolver:(RCTPromiseResolveBlock)resolve
+                 withRejecter:(RCTPromiseRejectBlock)reject)
+{
+    CactusContext *context = llamaContexts[[NSNumber numberWithDouble:contextId]];
+    if (context == nil) {
+        reject(@"llama_error", @"Context not found", nil);
+        return;
+    }
+    [context clearConversation];
+    resolve(nil);
+}
+
+RCT_EXPORT_METHOD(isConversationActive:(double)contextId
+                 withResolver:(RCTPromiseResolveBlock)resolve
+                 withRejecter:(RCTPromiseRejectBlock)reject)
+{
+    CactusContext *context = llamaContexts[[NSNumber numberWithDouble:contextId]];
+    if (context == nil) {
+        reject(@"llama_error", @"Context not found", nil);
+        return;
+    }
+    resolve(@([context isConversationActive]));
+}
+
 @end

package/ios/CactusContext.h

@@ -74,6 +74,12 @@
 - (NSArray *)decodeAudioTokens:(NSArray *)tokens;
 - (void)releaseVocoder;
 
+// New conversation management methods
+- (NSString *)generateResponse:(NSString *)userMessage maxTokens:(int)maxTokens;
+- (NSDictionary *)continueConversation:(NSString *)userMessage maxTokens:(int)maxTokens;
+- (void)clearConversation;
+- (BOOL)isConversationActive;
+
 - (void)invalidate;
 
 @end

package/ios/CactusContext.mm

@@ -961,6 +961,33 @@
     llama->releaseVocoder();
 }
 
+// New conversation management methods
+- (NSString *)generateResponse:(NSString *)userMessage maxTokens:(int)maxTokens {
+    std::string result = llama->generateResponse([userMessage UTF8String], maxTokens);
+    llama->is_predicting = false;
+    return [NSString stringWithUTF8String:result.c_str()];
+}
+
+- (NSDictionary *)continueConversation:(NSString *)userMessage maxTokens:(int)maxTokens {
+    cactus::conversation_result result = llama->continueConversation([userMessage UTF8String], maxTokens);
+    llama->is_predicting = false;
+    
+    return @{
+        @"text": [NSString stringWithUTF8String:result.text.c_str()],
+        @"time_to_first_token": @(result.time_to_first_token.count()),
+        @"total_time": @(result.total_time.count()),
+        @"tokens_generated": @(result.tokens_generated)
+    };
+}
+
+- (void)clearConversation {
+    llama->clearConversation();
+}
+
+- (BOOL)isConversationActive {
+    return llama->isConversationActive();
+}
+
 - (void)invalidate {
     delete llama;
     // llama_backend_free();

package/ios/cactus.xcframework/ios-arm64/cactus.framework/Headers/cactus.h

@@ -3,6 +3,7 @@
 
 #include <sstream>
 #include <iostream>
+#include <chrono>
 #include "chat.h"
 #include "common.h"
 #include "ggml.h"
@@ -48,6 +49,13 @@ struct completion_token_output
     llama_token tok;
 };
 
+struct conversation_result {
+    std::string text;
+    std::chrono::milliseconds time_to_first_token;
+    std::chrono::milliseconds total_time;
+    int tokens_generated;
+};
+
 struct cactus_tokenize_result {
     std::vector<llama_token> tokens;
     bool has_media = false;
@@ -112,6 +120,10 @@ struct cactus_context {
     bool has_vocoder = false;
     std::vector<llama_token> audio_tokens;
 
+    // Conversation management state
+    bool conversation_active = false;
+    std::string last_chat_template = "";
+
     ~cactus_context();
 
     void rewind();
@@ -180,6 +192,12 @@ struct cactus_context {
     std::vector<llama_token> getAudioCompletionGuideTokens(const std::string &text_to_speak);
     std::vector<float> decodeAudioTokens(const std::vector<llama_token> &tokens);
     void releaseVocoder();
+
+    // High-level conversation management API
+    std::string generateResponse(const std::string &user_message, int max_tokens = 200);
+    conversation_result continueConversation(const std::string &user_message, int max_tokens = 200);
+    void clearConversation();
+    bool isConversationActive() const;
 };
 
 extern bool cactus_verbose;

package/ios/cactus.xcframework/ios-arm64/cactus.framework/Headers/cactus_ffi.h

@@ -127,6 +127,15 @@ CACTUS_FFI_EXPORT int cactus_completion_c(
     cactus_completion_result_c_t* result
 );
 
+// **MULTIMODAL COMPLETION**
+CACTUS_FFI_EXPORT int cactus_multimodal_completion_c(
+    cactus_context_handle_t handle,
+    const cactus_completion_params_c_t* params,
+    const char** media_paths,
+    int media_count,
+    cactus_completion_result_c_t* result
+);
+
 CACTUS_FFI_EXPORT void cactus_stop_completion_c(cactus_context_handle_t handle);
 
 CACTUS_FFI_EXPORT cactus_token_array_c_t cactus_tokenize_c(cactus_context_handle_t handle, const char* text);
@@ -207,10 +216,39 @@ CACTUS_FFI_EXPORT cactus_lora_adapters_c_t cactus_get_loaded_lora_adapters_c(cac
 CACTUS_FFI_EXPORT bool cactus_validate_chat_template_c(cactus_context_handle_t handle, bool use_jinja, const char* name);
 CACTUS_FFI_EXPORT char* cactus_get_formatted_chat_c(cactus_context_handle_t handle, const char* messages, const char* chat_template);
 
+// **ADVANCED: Chat with Jinja and Tools Support**
+typedef struct {
+    char* prompt;
+    char* json_schema;
+    char* tools;
+    char* tool_choice;
+    bool parallel_tool_calls;
+} cactus_chat_result_c_t;
+
+CACTUS_FFI_EXPORT cactus_chat_result_c_t cactus_get_formatted_chat_with_jinja_c(
+    cactus_context_handle_t handle, 
+    const char* messages,
+    const char* chat_template,
+    const char* json_schema,
+    const char* tools,
+    bool parallel_tool_calls,
+    const char* tool_choice
+);
+
 // **HIGH PRIORITY: Context Management**
 CACTUS_FFI_EXPORT void cactus_rewind_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT bool cactus_init_sampling_c(cactus_context_handle_t handle);
 
+// **COMPLETION CONTROL**
+CACTUS_FFI_EXPORT void cactus_begin_completion_c(cactus_context_handle_t handle);
+CACTUS_FFI_EXPORT void cactus_end_completion_c(cactus_context_handle_t handle);
+CACTUS_FFI_EXPORT void cactus_load_prompt_c(cactus_context_handle_t handle);
+CACTUS_FFI_EXPORT void cactus_load_prompt_with_media_c(cactus_context_handle_t handle, const char** media_paths, int media_count);
+
+// **TOKEN PROCESSING**
+CACTUS_FFI_EXPORT int cactus_do_completion_step_c(cactus_context_handle_t handle, char** token_text);
+CACTUS_FFI_EXPORT size_t cactus_find_stopping_strings_c(cactus_context_handle_t handle, const char* text, size_t last_token_size, int stop_type);
+
 // **HIGH PRIORITY: Model Information**
 CACTUS_FFI_EXPORT int32_t cactus_get_n_ctx_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT int32_t cactus_get_n_embd_c(cactus_context_handle_t handle);
@@ -221,6 +259,7 @@ CACTUS_FFI_EXPORT int64_t cactus_get_model_params_c(cactus_context_handle_t hand
 // Memory management functions
 CACTUS_FFI_EXPORT void cactus_free_bench_result_members_c(cactus_bench_result_c_t* result);
 CACTUS_FFI_EXPORT void cactus_free_lora_adapters_c(cactus_lora_adapters_c_t* adapters);
+CACTUS_FFI_EXPORT void cactus_free_chat_result_members_c(cactus_chat_result_c_t* result);
 
 #ifdef __cplusplus
 }

package/ios/cactus.xcframework/ios-arm64_x86_64-simulator/cactus.framework/Headers/cactus.h

@@ -3,6 +3,7 @@
 
 #include <sstream>
 #include <iostream>
+#include <chrono>
 #include "chat.h"
 #include "common.h"
 #include "ggml.h"
@@ -48,6 +49,13 @@ struct completion_token_output
     llama_token tok;
 };
 
+struct conversation_result {
+    std::string text;
+    std::chrono::milliseconds time_to_first_token;
+    std::chrono::milliseconds total_time;
+    int tokens_generated;
+};
+
 struct cactus_tokenize_result {
     std::vector<llama_token> tokens;
     bool has_media = false;
@@ -112,6 +120,10 @@ struct cactus_context {
     bool has_vocoder = false;
     std::vector<llama_token> audio_tokens;
 
+    // Conversation management state
+    bool conversation_active = false;
+    std::string last_chat_template = "";
+
     ~cactus_context();
 
     void rewind();
@@ -180,6 +192,12 @@ struct cactus_context {
     std::vector<llama_token> getAudioCompletionGuideTokens(const std::string &text_to_speak);
     std::vector<float> decodeAudioTokens(const std::vector<llama_token> &tokens);
     void releaseVocoder();
+
+    // High-level conversation management API
+    std::string generateResponse(const std::string &user_message, int max_tokens = 200);
+    conversation_result continueConversation(const std::string &user_message, int max_tokens = 200);
+    void clearConversation();
+    bool isConversationActive() const;
 };
 
 extern bool cactus_verbose;

package/ios/cactus.xcframework/ios-arm64_x86_64-simulator/cactus.framework/Headers/cactus_ffi.h

@@ -127,6 +127,15 @@ CACTUS_FFI_EXPORT int cactus_completion_c(
     cactus_completion_result_c_t* result
 );
 
+// **MULTIMODAL COMPLETION**
+CACTUS_FFI_EXPORT int cactus_multimodal_completion_c(
+    cactus_context_handle_t handle,
+    const cactus_completion_params_c_t* params,
+    const char** media_paths,
+    int media_count,
+    cactus_completion_result_c_t* result
+);
+
 CACTUS_FFI_EXPORT void cactus_stop_completion_c(cactus_context_handle_t handle);
 
 CACTUS_FFI_EXPORT cactus_token_array_c_t cactus_tokenize_c(cactus_context_handle_t handle, const char* text);
@@ -207,10 +216,39 @@ CACTUS_FFI_EXPORT cactus_lora_adapters_c_t cactus_get_loaded_lora_adapters_c(cac
 CACTUS_FFI_EXPORT bool cactus_validate_chat_template_c(cactus_context_handle_t handle, bool use_jinja, const char* name);
 CACTUS_FFI_EXPORT char* cactus_get_formatted_chat_c(cactus_context_handle_t handle, const char* messages, const char* chat_template);
 
+// **ADVANCED: Chat with Jinja and Tools Support**
+typedef struct {
+    char* prompt;
+    char* json_schema;
+    char* tools;
+    char* tool_choice;
+    bool parallel_tool_calls;
+} cactus_chat_result_c_t;
+
+CACTUS_FFI_EXPORT cactus_chat_result_c_t cactus_get_formatted_chat_with_jinja_c(
+    cactus_context_handle_t handle, 
+    const char* messages,
+    const char* chat_template,
+    const char* json_schema,
+    const char* tools,
+    bool parallel_tool_calls,
+    const char* tool_choice
+);
+
 // **HIGH PRIORITY: Context Management**
 CACTUS_FFI_EXPORT void cactus_rewind_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT bool cactus_init_sampling_c(cactus_context_handle_t handle);
 
+// **COMPLETION CONTROL**
+CACTUS_FFI_EXPORT void cactus_begin_completion_c(cactus_context_handle_t handle);
+CACTUS_FFI_EXPORT void cactus_end_completion_c(cactus_context_handle_t handle);
+CACTUS_FFI_EXPORT void cactus_load_prompt_c(cactus_context_handle_t handle);
+CACTUS_FFI_EXPORT void cactus_load_prompt_with_media_c(cactus_context_handle_t handle, const char** media_paths, int media_count);
+
+// **TOKEN PROCESSING**
+CACTUS_FFI_EXPORT int cactus_do_completion_step_c(cactus_context_handle_t handle, char** token_text);
+CACTUS_FFI_EXPORT size_t cactus_find_stopping_strings_c(cactus_context_handle_t handle, const char* text, size_t last_token_size, int stop_type);
+
 // **HIGH PRIORITY: Model Information**
 CACTUS_FFI_EXPORT int32_t cactus_get_n_ctx_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT int32_t cactus_get_n_embd_c(cactus_context_handle_t handle);
@@ -221,6 +259,7 @@ CACTUS_FFI_EXPORT int64_t cactus_get_model_params_c(cactus_context_handle_t hand
 // Memory management functions
 CACTUS_FFI_EXPORT void cactus_free_bench_result_members_c(cactus_bench_result_c_t* result);
 CACTUS_FFI_EXPORT void cactus_free_lora_adapters_c(cactus_lora_adapters_c_t* adapters);
+CACTUS_FFI_EXPORT void cactus_free_chat_result_members_c(cactus_chat_result_c_t* result);
 
 #ifdef __cplusplus
 }

package/ios/cactus.xcframework/tvos-arm64/cactus.framework/Headers/cactus.h

@@ -3,6 +3,7 @@
 
 #include <sstream>
 #include <iostream>
+#include <chrono>
 #include "chat.h"
 #include "common.h"
 #include "ggml.h"
@@ -48,6 +49,13 @@ struct completion_token_output
     llama_token tok;
 };
 
+struct conversation_result {
+    std::string text;
+    std::chrono::milliseconds time_to_first_token;
+    std::chrono::milliseconds total_time;
+    int tokens_generated;
+};
+
 struct cactus_tokenize_result {
     std::vector<llama_token> tokens;
     bool has_media = false;
@@ -112,6 +120,10 @@ struct cactus_context {
     bool has_vocoder = false;
     std::vector<llama_token> audio_tokens;
 
+    // Conversation management state
+    bool conversation_active = false;
+    std::string last_chat_template = "";
+
     ~cactus_context();
 
     void rewind();
@@ -180,6 +192,12 @@ struct cactus_context {
     std::vector<llama_token> getAudioCompletionGuideTokens(const std::string &text_to_speak);
     std::vector<float> decodeAudioTokens(const std::vector<llama_token> &tokens);
     void releaseVocoder();
+
+    // High-level conversation management API
+    std::string generateResponse(const std::string &user_message, int max_tokens = 200);
+    conversation_result continueConversation(const std::string &user_message, int max_tokens = 200);
+    void clearConversation();
+    bool isConversationActive() const;
 };
 
 extern bool cactus_verbose;

package/ios/cactus.xcframework/tvos-arm64/cactus.framework/Headers/cactus_ffi.h

@@ -127,6 +127,15 @@ CACTUS_FFI_EXPORT int cactus_completion_c(
     cactus_completion_result_c_t* result
 );
 
+// **MULTIMODAL COMPLETION**
+CACTUS_FFI_EXPORT int cactus_multimodal_completion_c(
+    cactus_context_handle_t handle,
+    const cactus_completion_params_c_t* params,
+    const char** media_paths,
+    int media_count,
+    cactus_completion_result_c_t* result
+);
+
 CACTUS_FFI_EXPORT void cactus_stop_completion_c(cactus_context_handle_t handle);
 
 CACTUS_FFI_EXPORT cactus_token_array_c_t cactus_tokenize_c(cactus_context_handle_t handle, const char* text);
@@ -207,10 +216,39 @@ CACTUS_FFI_EXPORT cactus_lora_adapters_c_t cactus_get_loaded_lora_adapters_c(cac
 CACTUS_FFI_EXPORT bool cactus_validate_chat_template_c(cactus_context_handle_t handle, bool use_jinja, const char* name);
 CACTUS_FFI_EXPORT char* cactus_get_formatted_chat_c(cactus_context_handle_t handle, const char* messages, const char* chat_template);
 
+// **ADVANCED: Chat with Jinja and Tools Support**
+typedef struct {
+    char* prompt;
+    char* json_schema;
+    char* tools;
+    char* tool_choice;
+    bool parallel_tool_calls;
+} cactus_chat_result_c_t;
+
+CACTUS_FFI_EXPORT cactus_chat_result_c_t cactus_get_formatted_chat_with_jinja_c(
+    cactus_context_handle_t handle, 
+    const char* messages,
+    const char* chat_template,
+    const char* json_schema,
+    const char* tools,
+    bool parallel_tool_calls,
+    const char* tool_choice
+);
+
 // **HIGH PRIORITY: Context Management**
 CACTUS_FFI_EXPORT void cactus_rewind_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT bool cactus_init_sampling_c(cactus_context_handle_t handle);
 
+// **COMPLETION CONTROL**
+CACTUS_FFI_EXPORT void cactus_begin_completion_c(cactus_context_handle_t handle);
+CACTUS_FFI_EXPORT void cactus_end_completion_c(cactus_context_handle_t handle);
+CACTUS_FFI_EXPORT void cactus_load_prompt_c(cactus_context_handle_t handle);
+CACTUS_FFI_EXPORT void cactus_load_prompt_with_media_c(cactus_context_handle_t handle, const char** media_paths, int media_count);
+
+// **TOKEN PROCESSING**
+CACTUS_FFI_EXPORT int cactus_do_completion_step_c(cactus_context_handle_t handle, char** token_text);
+CACTUS_FFI_EXPORT size_t cactus_find_stopping_strings_c(cactus_context_handle_t handle, const char* text, size_t last_token_size, int stop_type);
+
 // **HIGH PRIORITY: Model Information**
 CACTUS_FFI_EXPORT int32_t cactus_get_n_ctx_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT int32_t cactus_get_n_embd_c(cactus_context_handle_t handle);
@@ -221,6 +259,7 @@ CACTUS_FFI_EXPORT int64_t cactus_get_model_params_c(cactus_context_handle_t hand
 // Memory management functions
 CACTUS_FFI_EXPORT void cactus_free_bench_result_members_c(cactus_bench_result_c_t* result);
 CACTUS_FFI_EXPORT void cactus_free_lora_adapters_c(cactus_lora_adapters_c_t* adapters);
+CACTUS_FFI_EXPORT void cactus_free_chat_result_members_c(cactus_chat_result_c_t* result);
 
 #ifdef __cplusplus
 }

package/ios/cactus.xcframework/tvos-arm64_x86_64-simulator/cactus.framework/Headers/cactus.h

@@ -3,6 +3,7 @@
 
 #include <sstream>
 #include <iostream>
+#include <chrono>
 #include "chat.h"
 #include "common.h"
 #include "ggml.h"
@@ -48,6 +49,13 @@ struct completion_token_output
     llama_token tok;
 };
 
+struct conversation_result {
+    std::string text;
+    std::chrono::milliseconds time_to_first_token;
+    std::chrono::milliseconds total_time;
+    int tokens_generated;
+};
+
 struct cactus_tokenize_result {
     std::vector<llama_token> tokens;
     bool has_media = false;
@@ -112,6 +120,10 @@ struct cactus_context {
     bool has_vocoder = false;
     std::vector<llama_token> audio_tokens;
 
+    // Conversation management state
+    bool conversation_active = false;
+    std::string last_chat_template = "";
+
     ~cactus_context();
 
     void rewind();
@@ -180,6 +192,12 @@ struct cactus_context {
     std::vector<llama_token> getAudioCompletionGuideTokens(const std::string &text_to_speak);
     std::vector<float> decodeAudioTokens(const std::vector<llama_token> &tokens);
     void releaseVocoder();
+
+    // High-level conversation management API
+    std::string generateResponse(const std::string &user_message, int max_tokens = 200);
+    conversation_result continueConversation(const std::string &user_message, int max_tokens = 200);
+    void clearConversation();
+    bool isConversationActive() const;
 };
 
 extern bool cactus_verbose;

package/ios/cactus.xcframework/tvos-arm64_x86_64-simulator/cactus.framework/Headers/cactus_ffi.h

@@ -127,6 +127,15 @@ CACTUS_FFI_EXPORT int cactus_completion_c(
     cactus_completion_result_c_t* result
 );
 
+// **MULTIMODAL COMPLETION**
+CACTUS_FFI_EXPORT int cactus_multimodal_completion_c(
+    cactus_context_handle_t handle,
+    const cactus_completion_params_c_t* params,
+    const char** media_paths,
+    int media_count,
+    cactus_completion_result_c_t* result
+);
+
 CACTUS_FFI_EXPORT void cactus_stop_completion_c(cactus_context_handle_t handle);
 
 CACTUS_FFI_EXPORT cactus_token_array_c_t cactus_tokenize_c(cactus_context_handle_t handle, const char* text);
@@ -207,10 +216,39 @@ CACTUS_FFI_EXPORT cactus_lora_adapters_c_t cactus_get_loaded_lora_adapters_c(cac
 CACTUS_FFI_EXPORT bool cactus_validate_chat_template_c(cactus_context_handle_t handle, bool use_jinja, const char* name);
 CACTUS_FFI_EXPORT char* cactus_get_formatted_chat_c(cactus_context_handle_t handle, const char* messages, const char* chat_template);
 
+// **ADVANCED: Chat with Jinja and Tools Support**
+typedef struct {
+    char* prompt;
+    char* json_schema;
+    char* tools;
+    char* tool_choice;
+    bool parallel_tool_calls;
+} cactus_chat_result_c_t;
+
+CACTUS_FFI_EXPORT cactus_chat_result_c_t cactus_get_formatted_chat_with_jinja_c(
+    cactus_context_handle_t handle, 
+    const char* messages,
+    const char* chat_template,
+    const char* json_schema,
+    const char* tools,
+    bool parallel_tool_calls,
+    const char* tool_choice
+);
+
 // **HIGH PRIORITY: Context Management**
 CACTUS_FFI_EXPORT void cactus_rewind_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT bool cactus_init_sampling_c(cactus_context_handle_t handle);
 
+// **COMPLETION CONTROL**
+CACTUS_FFI_EXPORT void cactus_begin_completion_c(cactus_context_handle_t handle);
+CACTUS_FFI_EXPORT void cactus_end_completion_c(cactus_context_handle_t handle);
+CACTUS_FFI_EXPORT void cactus_load_prompt_c(cactus_context_handle_t handle);
+CACTUS_FFI_EXPORT void cactus_load_prompt_with_media_c(cactus_context_handle_t handle, const char** media_paths, int media_count);
+
+// **TOKEN PROCESSING**
+CACTUS_FFI_EXPORT int cactus_do_completion_step_c(cactus_context_handle_t handle, char** token_text);
+CACTUS_FFI_EXPORT size_t cactus_find_stopping_strings_c(cactus_context_handle_t handle, const char* text, size_t last_token_size, int stop_type);
+
 // **HIGH PRIORITY: Model Information**
 CACTUS_FFI_EXPORT int32_t cactus_get_n_ctx_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT int32_t cactus_get_n_embd_c(cactus_context_handle_t handle);
@@ -221,6 +259,7 @@ CACTUS_FFI_EXPORT int64_t cactus_get_model_params_c(cactus_context_handle_t hand
 // Memory management functions
 CACTUS_FFI_EXPORT void cactus_free_bench_result_members_c(cactus_bench_result_c_t* result);
 CACTUS_FFI_EXPORT void cactus_free_lora_adapters_c(cactus_lora_adapters_c_t* adapters);
+CACTUS_FFI_EXPORT void cactus_free_chat_result_members_c(cactus_chat_result_c_t* result);
 
 #ifdef __cplusplus
 }

package/lib/commonjs/NativeCactus.js

@@ -6,5 +6,6 @@ Object.defineProperty(exports, "__esModule", {
 exports.default = void 0;
 var _reactNative = require("react-native");
 // New TTS/Audio types
+// New conversation management types
 var _default = exports.default = _reactNative.TurboModuleRegistry.get('Cactus');
 //# sourceMappingURL=NativeCactus.js.map

package/lib/commonjs/NativeCactus.js.map

@@ -1 +1 @@
-{"version":3,"names":["_reactNative","require","_default","exports","default","TurboModuleRegistry","get"],"sourceRoot":"../../src","sources":["NativeCactus.ts"],"mappings":";;;;;;AACA,IAAAA,YAAA,GAAAC,OAAA;AAkSA;AAAA,IAAAC,QAAA,GAAAC,OAAA,CAAAC,OAAA,GA2LeC,gCAAmB,CAACC,GAAG,CAAO,QAAQ,CAAC","ignoreList":[]}
+{"version":3,"names":["_reactNative","require","_default","exports","default","TurboModuleRegistry","get"],"sourceRoot":"../../src","sources":["NativeCactus.ts"],"mappings":";;;;;;AACA,IAAAA,YAAA,GAAAC,OAAA;AAkSA;AAiBA;AAAA,IAAAC,QAAA,GAAAC,OAAA,CAAAC,OAAA,GAgMeC,gCAAmB,CAACC,GAAG,CAAO,QAAQ,CAAC","ignoreList":[]}

package/lib/commonjs/index.js

@@ -216,6 +216,35 @@ class LlamaContext {
   stopCompletion() {
     return _NativeCactus.default.stopCompletion(this.id);
   }
+
+  // New conversation management methods
+  /**
+   * Generate a response to a user message (simple text return)
+   */
+  async generateResponse(userMessage, maxTokens = 200) {
+    return _NativeCactus.default.generateResponse(this.id, userMessage, maxTokens);
+  }
+
+  /**
+   * Continue conversation with detailed timing information
+   */
+  async continueConversation(userMessage, maxTokens = 200) {
+    return _NativeCactus.default.continueConversation(this.id, userMessage, maxTokens);
+  }
+
+  /**
+   * Clear conversation history and reset KV cache
+   */
+  async clearConversation() {
+    return _NativeCactus.default.clearConversation(this.id);
+  }
+
+  /**
+   * Check if conversation is currently active
+   */
+  async isConversationActive() {
+    return _NativeCactus.default.isConversationActive(this.id);
+  }
   tokenize(text) {
     return _NativeCactus.default.tokenize(this.id, text);
   }