add computer, bash, text editor tools

.gitmodules

@@ -1,3 +0,0 @@
-[submodule "stream/thirdparty/llama.cpp"]
-	path = stream/thirdparty/llama.cpp
-	url = https://github.com/ggml-org/llama.cpp

README.md

@@ -0,0 +1 @@
+# computeruse

computeruse/README.md

@@ -1,2 +0,0 @@
-# computeruse
-This is a prototype.

computeruse/quartz_doubleclick.py

@@ -1,39 +0,0 @@
-'''
-This script simulates a double-click at the given mouse cursor position.
-'''
-
-import Quartz
-from time import sleep
-
-def post_mouse_event(type, pos, click_state):
-    event = Quartz.CGEventCreateMouseEvent(
-        None, type, pos, Quartz.kCGMouseButtonLeft
-    )
-    Quartz.CGEventSetIntegerValueField(event, Quartz.kCGMouseEventClickState, click_state)
-    Quartz.CGEventPost(Quartz.kCGHIDEventTap, event)
-
-def double_click (x, y):
-    pos = None
-    if x is None or y is None:
-        # Get current mouse position if no coordinates are provided
-        loc = Quartz.CGEventGetLocation(Quartz.CGEventCreate(None))
-        pos = (loc.x, loc.y)
-    else:
-        # Use provided coordinates
-        pos = (x, y)
-
-    # First click
-    post_mouse_event(Quartz.kCGEventLeftMouseDown, pos, 1)
-    post_mouse_event(Quartz.kCGEventLeftMouseUp, pos, 1)
-
-    sleep(0.05)  # Short delay within double-click threshold
-
-    # Second click
-    post_mouse_event(Quartz.kCGEventLeftMouseDown, pos, 2)
-    post_mouse_event(Quartz.kCGEventLeftMouseUp, pos, 2)
-
-
-if __name__ == "__main__":
-    print('test: double clicking at cursor position in 3 seconds...')
-    sleep(3)
-    double_click()

main.py

@@ -6,14 +6,11 @@ import pyautogui
 import subprocess
 from io import BytesIO
 from PIL import Image
-import quartz_doubleclick
 
 ANTHROPIC_API_URL = "https://api.anthropic.com/v1/messages"
 MODEL = "claude-3-7-sonnet-20250219"
 BETA_FLAG = "computer-use-2025-01-24"
 TOOL_VERSION = "20250124"
-pyautogui.PAUSE = 0.1
-SCALING_FACTOR = 1.25
 
 HEADERS = {
     "content-type": "application/json",
@@ -51,9 +48,8 @@ def execute_computer_tool(tool_input):
         buffered = BytesIO()
         screenshot.save(buffered, format="PNG")
 
-        MAX_BINARY_SIZE = 5242880 * 3 // 4
-        img_data = None
-        if buffered.tell() > MAX_BINARY_SIZE:
+        # Check size and resize if needed (5MB = 5242880 bytes)
+        if buffered.tell() > 4242880:
             # Reset buffer
             buffered.seek(0)
             screenshot = Image.open(buffered)
@@ -62,19 +58,14 @@ def execute_computer_tool(tool_input):
             while True:
                 buffered = BytesIO()
                 new_size = (int(screenshot.width * 0.8), int(screenshot.height * 0.8))
-                print('new size:', new_size)
                 screenshot = screenshot.resize(new_size, Image.Resampling.LANCZOS)
                 screenshot.save(buffered, format="PNG", optimize=True)
-                buffered.seek(0)
-                img_data = buffered.read()
-                if len(img_data) <= MAX_BINARY_SIZE:
+                if buffered.tell() <= 4242880:
                     break
-                buffered.seek(0)
-                screenshot = Image.open(buffered)
-        else:
-            img_data = buffered.getvalue()
+         
         # Convert to base64
-        img_base64 = base64.b64encode(img_data).decode('utf-8')
+        buffered.seek(0)
+        img_base64 = base64.b64encode(buffered.read()).decode('utf-8')
         
         return {
             "type": "image",
@@ -85,10 +76,10 @@ def execute_computer_tool(tool_input):
             }
         }
     
-    elif action == "left_click":
+    elif action == "click":
         # Get coordinates
-        x = tool_input.get("coordinate")[0] * SCALING_FACTOR
-        y = tool_input.get("coordinate")[1] * SCALING_FACTOR
+        x = tool_input.get("coordinate_x")
+        y = tool_input.get("coordinate_y")
         if x is None or y is None:
             return {"type": "text", "text": "Error: Missing coordinates for click action"}
         
@@ -98,14 +89,13 @@ def execute_computer_tool(tool_input):
     
     elif action == "double_click":
         # Get coordinates
-        x = tool_input.get("coordinate")[0] * SCALING_FACTOR
-        y = tool_input.get("coordinate")[1] * SCALING_FACTOR
+        x = tool_input.get("coordinate_x")
+        y = tool_input.get("coordinate_y")
         if x is None or y is None:
             return {"type": "text", "text": "Error: Missing coordinates for double click action"}
         
         # Perform double click
-        # pyautogui.doubleClick(x, y, interval=0.2) this doesn't work
-        quartz_doubleclick.double_click(x, y)
+        pyautogui.doubleClick(x, y)
         return {"type": "text", "text": f"Double-clicked at coordinates ({x}, {y})"}
     
     elif action == "type":
@@ -116,24 +106,19 @@ def execute_computer_tool(tool_input):
     
     elif action == "key":
         # Press a key or key combination
-        text = tool_input.get("text", "")
+        key = tool_input.get("key", "")
         try:
-            if '+' in text:
-                # Handle key combinations like "command+c"
-                keys = text.replace('super', 'command').split('+')
-                pyautogui.hotkey(*keys, interval=0.05) # interval is required
-            else:
-                pyautogui.press(text)
-            return {"type": "text", "text": f"Pressed key: {text}"}
+            pyautogui.press(key)
+            return {"type": "text", "text": f"Pressed key: {key}"}
         except Exception as e:
-            return {"type": "text", "text": f"Error pressing key {text}: {str(e)}"}
+            return {"type": "text", "text": f"Error pressing key {key}: {str(e)}"}
     
     elif action == "scroll":
-        # Scroll action (should we really have defaults here?)
-        direction = tool_input.get("scroll_direction", "down")
-        amount = tool_input.get("scroll_amount", 3)
+        # Scroll action
+        direction = tool_input.get("direction", "down")
+        amount = tool_input.get("amount", 3)
         
-        scroll_amount = -amount if direction == "down" else amount
+        scroll_amount = -amount if direction == "up" else amount
         pyautogui.scroll(scroll_amount)
         return {"type": "text", "text": f"Scrolled {direction} by {amount}"}
     
@@ -165,23 +150,23 @@ def execute_bash_tool(command):
         return {"type": "text", "text": f"Error executing command: {str(e)}"}
 
 
-def execute_text_editor_tool(_command, _path, **kwargs):
+def execute_text_editor_tool(command, path, **kwargs):
     """Execute text editor actions."""
     
-    if _command == "view":
+    if command == "view":
         try:
-            with open(_path, 'r') as f:
+            with open(path, 'r') as f:
                 content = f.read()
             return {"type": "text", "text": content}
         except Exception as e:
             return {"type": "text", "text": f"Error reading file: {str(e)}"}
     
-    elif _command == "str_replace":
+    elif command == "str_replace":
         old_str = kwargs.get("old_str", "")
         new_str = kwargs.get("new_str", "")
         
         try:
-            with open(_path, 'r') as f:
+            with open(path, 'r') as f:
                 content = f.read()
             
             if old_str not in content:
@@ -191,7 +176,7 @@ def execute_text_editor_tool(_command, _path, **kwargs):
             new_content = content.replace(old_str, new_str)
             
             # Write back to file
-            with open(_path, 'w') as f:
+            with open(path, 'w') as f:
                 f.write(new_content)
             
             return {"type": "text", "text": "String replaced successfully"}
@@ -199,17 +184,17 @@ def execute_text_editor_tool(_command, _path, **kwargs):
         except Exception as e:
             return {"type": "text", "text": f"Error modifying file: {str(e)}"}
     
-    elif _command == "create":
+    elif command == "create":
         content = kwargs.get("content", "")
         try:
-            with open(_path, 'w') as f:
+            with open(path, 'w') as f:
                 f.write(content)
-            return {"type": "text", "text": f"File created: {_path}"}
+            return {"type": "text", "text": f"File created: {path}"}
         except Exception as e:
             return {"type": "text", "text": f"Error creating file: {str(e)}"}
     
     else:
-        return {"type": "text", "text": f"Unknown text editor command: {_command}"}
+        return {"type": "text", "text": f"Unknown text editor command: {command}"}
 
 
 def execute_tool(tool_name, tool_input):
@@ -246,8 +231,8 @@ def send_prompt(api_key, messages):
             {
                 "type": f"computer_{TOOL_VERSION}",
                 "name": "computer",
-                "display_width_px": 2880,
-                "display_height_px": 1864,
+                "display_width_px": 1024,
+                "display_height_px": 768,
                 "display_number": 1
             },
             {
@@ -330,12 +315,6 @@ def run():
             
             # Execute the tool
             result_content = execute_tool(tool_name, tool_input)
-            print('[TOOL RESULT]')
-            for key in result_content: 
-                if key == "source":
-                    print(f"{key}: {len(result_content[key])} chars long")
-                else:
-                    print(f"{key}: {result_content[key]}")
             
             # Add to results
             tool_results.append({

stream/.gitignore

@@ -1,4 +0,0 @@
-.cache
-stream_app
-*.gguf
-compile_commands.json

stream/Makefile

@@ -1,11 +0,0 @@
-CC = gcc
-CFLAGS = -Wall -g -I./thirdparty/llama.cpp/include -I./thirdparty/llama.cpp/ggml/include
-TARGET = stream_app
-SRC = main.c
-LIBS = -L./thirdparty/llama.cpp/build/bin -lllama -lggml -lggml-base
-
-$(TARGET): $(SRC)
-	$(CC) $(CFLAGS) -o $(TARGET) $(SRC) $(LIBS)
-
-clean:
-	rm -f *.o $(TARGET)

stream/README.md

@@ -1,38 +0,0 @@
-# flowy.stream
-
-This is a proof of concept of a showing widgets within a conversation based on intent.
-
-## Todo
-- [x] Integrate llama.cpp with local inference. This will set us up for building many parts of experience.
-- [ ] Disect what llama is doing and what the Phi model is doing.
-- [ ] Create conversational loop with chat and running context.
-- [ ] Generate & render different types of blocks: list, email, doc, etc.
-- [ ] Try different models with hugging face
-- [ ] Render TUI elements based on commands
-
-## Dependencies
-
-### llama.cpp
-```sh
-#model weights
-wget https://huggingface.co/microsoft/Phi-3-mini-4k-instruct-gguf/resolve/main/Phi-3-mini-4k-instruct-q4.gguf
-```
-
-```sh
-cd ./thirdparty/llama.cpp
-
-rm -rf build
-mkdir -p build
-cmake -S. -Bbuild -DLLAMA_CURL=OFF
-cmake --build build
-```
-
-The libraries should be in `./thirdparty/llama.cpp/build/bin`.
-
-NOTE: you may have to disable curl as a flag when configuring llama.cpp via cmake.
-
-## Getting started
-```sh
-make
-LD_LIBRARY_PATH=./thirdparty/llama.cpp/build/bin/ ./stream_app
-```

stream/main.c

@@ -1,176 +0,0 @@
-#include "thirdparty/llama.cpp/ggml/include/ggml-backend.h"
-#include <llama.h>
-#include <stdio.h>
-#include <stdlib.h>
-#include <string.h>
-
-typedef enum {
-  Unspecified,
-  Question,
-  Math,
-  Code,
-  Email,
-  Doc,
-  WebSearch,
-} IntentType;
-
-int main(int argc, char *argv[]) {
-  printf("Hello world\n");
-
-  char user_input[500];
-  printf("What's up, what can I help with?\n");
-  fgets(user_input, sizeof(user_input), stdin);
-
-  char prompt[1000];
-  snprintf(prompt, sizeof(prompt), "<|system|>You are a helpful assistant. Your goal is to take in what the user is doing and return 3 predictive actions/3 suggestions based on what the user is trying to do: e.g. change page title when in google sheets, or calculate sum, create chart.<|end|>\n<|user|>%s<|end|>\n<|assistant|>", user_input);
-
-  // number of layers to offload to the GPU
-  int ngl = 99;
-  // number of tokens to predict
-  int n_predict = 1000;
-
-  // load dynamic backends
-
-  ggml_backend_load_all();
-
-  // initialize the model
-
-  struct llama_model_params model_params = llama_model_default_params();
-  model_params.n_gpu_layers = ngl;
-
-  struct llama_model *model = llama_model_load_from_file(
-      "Phi-3-mini-4k-instruct-q4.gguf", model_params);
-
-  if (model == NULL) {
-    fprintf(stderr, "%s: error: unable to load model\n", __func__);
-    return 1;
-  }
-
-  const struct llama_vocab *vocab = llama_model_get_vocab(model);
-  // tokenize the prompt
-
-  // find the number of tokens in the prompt
-  const int n_prompt =
-      -llama_tokenize(vocab, prompt, strlen(prompt), NULL, 0, true, true);
-
-  // allocate space for the tokens and tokenize the prompt
-  llama_token *prompt_tokens = malloc(n_prompt * sizeof(llama_token));
-  if (prompt_tokens == NULL) {
-    fprintf(stderr, "%s: error: failed to allocate memory for prompt tokens\n",
-            __func__);
-    return 1;
-  }
-  if (llama_tokenize(vocab, prompt, strlen(prompt), prompt_tokens, n_prompt,
-                     true, true) < 0) {
-    fprintf(stderr, "%s: error: failed to tokenize the prompt\n", __func__);
-    return 1;
-  }
-
-  // initialize the context
-
-  struct llama_context_params ctx_params = llama_context_default_params();
-  // n_ctx is the context size
-  ctx_params.n_ctx = n_prompt + n_predict - 1;
-  // n_batch is the maximum number of tokens that can be processed in a single
-  // call to llama_decode
-  ctx_params.n_batch = n_prompt;
-  // enable performance counters
-  ctx_params.no_perf = false;
-
-  struct llama_context *ctx = llama_init_from_model(model, ctx_params);
-
-  if (ctx == NULL) {
-    fprintf(stderr, "%s: error: failed to create the llama_context\n",
-            __func__);
-    return 1;
-  }
-
-  // initialize the sampler
-
-  struct llama_sampler_chain_params sparams =
-      llama_sampler_chain_default_params();
-  sparams.no_perf = false;
-  struct llama_sampler *smpl = llama_sampler_chain_init(sparams);
-
-  llama_sampler_chain_add(smpl, llama_sampler_init_greedy());
-
-  // print the prompt token-by-token
-
-  for (int i = 0; i < n_prompt; i++) {
-    char buf[128];
-    int n = llama_token_to_piece(vocab, prompt_tokens[i], buf, sizeof(buf), 0,
-                                 true);
-    if (n < 0) {
-      fprintf(stderr, "%s: error: failed to convert token to piece\n",
-              __func__);
-      return 1;
-    }
-    buf[n] = '\0';
-    printf("%s", buf);
-  }
-
-  // prepare a batch for the prompt
-
-  llama_batch batch = llama_batch_get_one(prompt_tokens, n_prompt);
-
-  // main loop
-
-  const int64_t t_main_start = ggml_time_us();
-  int n_decode = 0;
-  llama_token new_token_id;
-
-  for (int n_pos = 0; n_pos + batch.n_tokens < n_prompt + n_predict;) {
-    // evaluate the current batch with the transformer model
-    if (llama_decode(ctx, batch)) {
-      fprintf(stderr, "%s : failed to eval, return code %d\n", __func__, 1);
-      return 1;
-    }
-
-    n_pos += batch.n_tokens;
-
-    // sample the next token
-    {
-      new_token_id = llama_sampler_sample(smpl, ctx, -1);
-
-      // is it an end of generation?
-      if (llama_vocab_is_eog(vocab, new_token_id)) {
-        break;
-      }
-
-      char buf[128];
-      int n =
-          llama_token_to_piece(vocab, new_token_id, buf, sizeof(buf), 0, true);
-      if (n < 0) {
-        fprintf(stderr, "%s: error: failed to convert token to piece\n",
-                __func__);
-        return 1;
-      }
-      buf[n] = '\0';
-      printf("%s", buf);
-      fflush(stdout);
-
-      // prepare the next batch with the sampled token
-      batch = llama_batch_get_one(&new_token_id, 1);
-
-      n_decode += 1;
-    }
-  }
-
-  printf("\n");
-
-  const int64_t t_main_end = ggml_time_us();
-
-  fprintf(stderr, "%s: decoded %d tokens in %.2f s, speed: %.2f t/s\n",
-          __func__, n_decode, (t_main_end - t_main_start) / 1000000.0f,
-          n_decode / ((t_main_end - t_main_start) / 1000000.0f));
-
-  fprintf(stderr, "\n");
-  llama_perf_sampler_print(smpl);
-  llama_perf_context_print(ctx);
-  fprintf(stderr, "\n");
-
-  llama_sampler_free(smpl);
-  llama_free(ctx);
-  llama_model_free(model);
-  free(prompt_tokens);
-}