mlx-code 0.0.26__tar.gz → 0.0.28__tar.gz

{mlx_code-0.0.26 → mlx_code-0.0.28}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: mlx-code
-Version: 0.0.26
+Version: 0.0.28
 Summary: Coding Agent for Mac
 Home-page: https://josefalbers.github.io/mlx-code/
 Author: J Joe
@@ -40,16 +40,16 @@ Dynamic: summary
 
 A Git-native coding agent that can run entirely on your Mac. No API keys, no cloud, and no data leaving your machine. Powered by Apple MLX, it turns commits, branches, and worktrees into the agent’s state, history, and execution model
 
-[![Link](https://raw.githubusercontent.com/JosefAlbers/mlx-code/main/assets/mlx-code-v0.0.20.gif)](https://youtu.be/0lkY7YQCyCo)
+[![v0.0.27](https://github.com/user-attachments/assets/8a1c131a-dda1-4b52-9fa6-9c0fbccb5ea6)](https://youtube.com/shorts/1LuifKFKixc)
 
 ---
 
 ## Architecture
 
 ```
-Conversation tree (nodes = git commits with embedded chat history)
+Worktrees:
 
-  main ──●──●──●──●──●──●──●──●──●──●──●──●──●──●
+  main ──●──●──●──●──●──●──●──●──●──●──●──●──●──●───────────► Node = git commit + chat hx
             │        │
             │        └── branch-1 ──●──●──●
             │                          │ ┌────────────┐
@@ -57,31 +57,29 @@ Conversation tree (nodes = git commits with embedded chat history)
             │                            └─────┬──────┘
             └── branch-0 ──●──●──●             │
                                                │
+Tabs:                                          ├────────────► Tab = git branch + Agent
                                                │
-REPL tabs (each tab = a git branch + agent)    │
-                                               │
-                                               │
-┌──────────────────────────────────────────────┼─────────┐
+┌──────────────────────────────────────────────│─────────┐
 │  TUI tabs                                    │         │
 │  ┌──────┐  ┌──────────┐  ┌──────────┐  ┌─────┴──────┐  │
 │  │ main │  │ branch-0 │  │ branch-1 │  │ branch-1-0 │  │
 │  └──────┘  └────┬─────┘  └──────────┘  └────────────┘  │
-└─────────────────┼──────────────────────────────────────┘
+└─────────────────│──────────────────────────────────────┘
                   │
-                  ├─────────────────────────────────────────► Each tab is an independent Agent
+Agents:           ├─────────────────────────────────────────► Each tab runs its own Agent
                   │
              ┌────┴─────────────────────────────────────┐
              │  Agent                                   │
              │  ┌────────────────┐  ┌────────────────┐  │
              │  │ API:           │  │ Tools:         │  │
              │  │ Local (mlx-lm) │  │ Read    Write  │  │
-             │  │ Claude         │  │ Edit    Bash   │  │
-             │  │ Gemini         │  │ Grep    Find   │  │
-             │  │ OpenAI         │  │ Ls      Skill  │  │
-             │  └────────────────┘  │ Agent ─────────┼──┼───► Spawns child Agent
-             │                      └────────────────┘  │     (each with own tools + worktree + etc)
-             │  Git worktree                            │  
-             │  (isolation + session state)             │ 
+             │  │ Gemini         │  │ Edit    Bash   │  │
+             │  │ Claude         │  │ Grep    Find   │  │
+             │  │ Codex          │  │ Ls      Skill  │  │
+             │  │ DeepSeek       │  │ Agent ─────────┼──┼───► Recursively spawns sub-Agents
+             │  └────────────────┘  └────────────────┘  │     
+             │  Git worktree                            │
+             │  (isolation + session state)             │
              └──────────────────────────────────────────┘
 ```
 
@@ -97,6 +95,15 @@ result = await agent.run('refactor utils.py to use dataclasses')
 
 ---
 
+## Core ideas
+
+- **Git is the state machine.** Every file-changing agent step is committed with the conversation that produced it, so you can inspect, resume, and branch from any checkpoint.
+- **Branches are alternative futures.** A branch is not just a Git branch; it is a different reasoning path with its own worktree and session state.
+- **Agents are the primitive.** Tabs, branches, and delegated subtasks are all instances of the same `Agent` abstraction.
+- **Worktrees provide isolation.** The agent edits in a separate worktree, so your main checkout stays clean and recoverable.
+
+---
+
 ## Quick start
 
 ```bash
@@ -115,15 +122,6 @@ That's it. The first run starts a local inference server and drops you into the
 
 ---
 
-## Core ideas
-
-- **Git is the state machine.** Every file-changing agent step is committed with the conversation that produced it, so you can inspect, resume, and branch from any checkpoint.
-- **Branches are alternative futures.** A branch is not just a Git branch; it is a different reasoning path with its own worktree and session state.
-- **Agents are the primitive.** Tabs, branches, and delegated subtasks are all instances of the same `Agent` abstraction.
-- **Worktrees provide isolation.** The agent edits in a separate worktree, so your main checkout stays clean and recoverable.
-
----
-
 ## Why mlx-code
 
 **Agents as reusable workflow atoms.** Tabs, branches, and tasks are all managed within instances of `Agent`. Each one gets its own conversation, its own tools, and its own worktree. Agents can spawn sub-agents to delegate subtasks, and each child is a full agent with its own scoped tool set.
@@ -522,10 +520,10 @@ All file tools enforce path sandboxing. The agent cannot read or write outside t
 
 | Backend | Flag | Notes |
 |---------|------|-------|
-| MLX (local) | `--api noapi` | Default. Runs on-device, no API key needed |
+| MLX-LM (local) | `--api noapi` | Default. Runs on-device, no API key needed |
 | Claude | `--api claude` | Requires `ANTHROPIC_API_KEY` |
 | Gemini | `--api gemini` | Requires `GOOGLE_API_KEY` |
-| DeepSeek | `--api deepseek` | DeepSeek API or compatible endpoint |
+| DeepSeek | `--api deepseek` | Requires `DEEPSEEK_API_KEY` |
 | Codex | `--api codex` | OpenAI Codex CLI integration |
 | OpenAI | `--api openai` | Any OpenAI-compatible endpoint |
 
@@ -534,10 +532,25 @@ All file tools enforce path sandboxing. The agent cannot read or write outside t
 The local MLX server speaks OpenAI, Anthropic, and Gemini wire formats simultaneously, so you can use any compatible CLI as the frontend:
 
 ```bash
-mlc --leash claude       # claude CLI routes through local model
-mlc --leash codex        # codex CLI routes through local model
-mlc --leash gemini       # gemini CLI routes through local model
-mlc --leash none         # server only
+mlc                      # default
+mlc --web                # web UI (api.mlx-code.com)
+mlc --bare               # no TUI
+mlc --leash none         # no harness
+mlc --leash codex        # codex CLI
+mlc --leash gemini       # gemini CLI
+mlc --leash claude       # claude code
+```
+
+#### WebUI
+
+```bash
+[protect & connect]-[networking]-[tunnels]-[add route]-[add published application]:
+    - subdomain: jjoe
+    - domain: mlx-code.com
+    - service url: http://host.containers.internal:8080
+mlc --host 0.0.0.0 --engine batch --web &
+podman run cloudflare/cloudflared:latest tunnel --no-autoupdate run --token $JJ_CFD_TOKEN
+phone http://jjoe.mlx-code.com
 ```
 
 ---

{mlx_code-0.0.26 → mlx_code-0.0.28}/README.md

@@ -2,16 +2,16 @@
 
 A Git-native coding agent that can run entirely on your Mac. No API keys, no cloud, and no data leaving your machine. Powered by Apple MLX, it turns commits, branches, and worktrees into the agent’s state, history, and execution model
 
-[![Link](https://raw.githubusercontent.com/JosefAlbers/mlx-code/main/assets/mlx-code-v0.0.20.gif)](https://youtu.be/0lkY7YQCyCo)
+[![v0.0.27](https://github.com/user-attachments/assets/8a1c131a-dda1-4b52-9fa6-9c0fbccb5ea6)](https://youtube.com/shorts/1LuifKFKixc)
 
 ---
 
 ## Architecture
 
 ```
-Conversation tree (nodes = git commits with embedded chat history)
+Worktrees:
 
-  main ──●──●──●──●──●──●──●──●──●──●──●──●──●──●
+  main ──●──●──●──●──●──●──●──●──●──●──●──●──●──●───────────► Node = git commit + chat hx
             │        │
             │        └── branch-1 ──●──●──●
             │                          │ ┌────────────┐
@@ -19,31 +19,29 @@ Conversation tree (nodes = git commits with embedded chat history)
             │                            └─────┬──────┘
             └── branch-0 ──●──●──●             │
                                                │
+Tabs:                                          ├────────────► Tab = git branch + Agent
                                                │
-REPL tabs (each tab = a git branch + agent)    │
-                                               │
-                                               │
-┌──────────────────────────────────────────────┼─────────┐
+┌──────────────────────────────────────────────│─────────┐
 │  TUI tabs                                    │         │
 │  ┌──────┐  ┌──────────┐  ┌──────────┐  ┌─────┴──────┐  │
 │  │ main │  │ branch-0 │  │ branch-1 │  │ branch-1-0 │  │
 │  └──────┘  └────┬─────┘  └──────────┘  └────────────┘  │
-└─────────────────┼──────────────────────────────────────┘
+└─────────────────│──────────────────────────────────────┘
                   │
-                  ├─────────────────────────────────────────► Each tab is an independent Agent
+Agents:           ├─────────────────────────────────────────► Each tab runs its own Agent
                   │
              ┌────┴─────────────────────────────────────┐
              │  Agent                                   │
              │  ┌────────────────┐  ┌────────────────┐  │
              │  │ API:           │  │ Tools:         │  │
              │  │ Local (mlx-lm) │  │ Read    Write  │  │
-             │  │ Claude         │  │ Edit    Bash   │  │
-             │  │ Gemini         │  │ Grep    Find   │  │
-             │  │ OpenAI         │  │ Ls      Skill  │  │
-             │  └────────────────┘  │ Agent ─────────┼──┼───► Spawns child Agent
-             │                      └────────────────┘  │     (each with own tools + worktree + etc)
-             │  Git worktree                            │  
-             │  (isolation + session state)             │ 
+             │  │ Gemini         │  │ Edit    Bash   │  │
+             │  │ Claude         │  │ Grep    Find   │  │
+             │  │ Codex          │  │ Ls      Skill  │  │
+             │  │ DeepSeek       │  │ Agent ─────────┼──┼───► Recursively spawns sub-Agents
+             │  └────────────────┘  └────────────────┘  │     
+             │  Git worktree                            │
+             │  (isolation + session state)             │
              └──────────────────────────────────────────┘
 ```
 
@@ -59,6 +57,15 @@ result = await agent.run('refactor utils.py to use dataclasses')
 
 ---
 
+## Core ideas
+
+- **Git is the state machine.** Every file-changing agent step is committed with the conversation that produced it, so you can inspect, resume, and branch from any checkpoint.
+- **Branches are alternative futures.** A branch is not just a Git branch; it is a different reasoning path with its own worktree and session state.
+- **Agents are the primitive.** Tabs, branches, and delegated subtasks are all instances of the same `Agent` abstraction.
+- **Worktrees provide isolation.** The agent edits in a separate worktree, so your main checkout stays clean and recoverable.
+
+---
+
 ## Quick start
 
 ```bash
@@ -77,15 +84,6 @@ That's it. The first run starts a local inference server and drops you into the
 
 ---
 
-## Core ideas
-
-- **Git is the state machine.** Every file-changing agent step is committed with the conversation that produced it, so you can inspect, resume, and branch from any checkpoint.
-- **Branches are alternative futures.** A branch is not just a Git branch; it is a different reasoning path with its own worktree and session state.
-- **Agents are the primitive.** Tabs, branches, and delegated subtasks are all instances of the same `Agent` abstraction.
-- **Worktrees provide isolation.** The agent edits in a separate worktree, so your main checkout stays clean and recoverable.
-
----
-
 ## Why mlx-code
 
 **Agents as reusable workflow atoms.** Tabs, branches, and tasks are all managed within instances of `Agent`. Each one gets its own conversation, its own tools, and its own worktree. Agents can spawn sub-agents to delegate subtasks, and each child is a full agent with its own scoped tool set.
@@ -484,10 +482,10 @@ All file tools enforce path sandboxing. The agent cannot read or write outside t
 
 | Backend | Flag | Notes |
 |---------|------|-------|
-| MLX (local) | `--api noapi` | Default. Runs on-device, no API key needed |
+| MLX-LM (local) | `--api noapi` | Default. Runs on-device, no API key needed |
 | Claude | `--api claude` | Requires `ANTHROPIC_API_KEY` |
 | Gemini | `--api gemini` | Requires `GOOGLE_API_KEY` |
-| DeepSeek | `--api deepseek` | DeepSeek API or compatible endpoint |
+| DeepSeek | `--api deepseek` | Requires `DEEPSEEK_API_KEY` |
 | Codex | `--api codex` | OpenAI Codex CLI integration |
 | OpenAI | `--api openai` | Any OpenAI-compatible endpoint |
 
@@ -496,10 +494,25 @@ All file tools enforce path sandboxing. The agent cannot read or write outside t
 The local MLX server speaks OpenAI, Anthropic, and Gemini wire formats simultaneously, so you can use any compatible CLI as the frontend:
 
 ```bash
-mlc --leash claude       # claude CLI routes through local model
-mlc --leash codex        # codex CLI routes through local model
-mlc --leash gemini       # gemini CLI routes through local model
-mlc --leash none         # server only
+mlc                      # default
+mlc --web                # web UI (api.mlx-code.com)
+mlc --bare               # no TUI
+mlc --leash none         # no harness
+mlc --leash codex        # codex CLI
+mlc --leash gemini       # gemini CLI
+mlc --leash claude       # claude code
+```
+
+#### WebUI
+
+```bash
+[protect & connect]-[networking]-[tunnels]-[add route]-[add published application]:
+    - subdomain: jjoe
+    - domain: mlx-code.com
+    - service url: http://host.containers.internal:8080
+mlc --host 0.0.0.0 --engine batch --web &
+podman run cloudflare/cloudflared:latest tunnel --no-autoupdate run --token $JJ_CFD_TOKEN
+phone http://jjoe.mlx-code.com
 ```
 
 ---

mlx_code-0.0.28/mlx_code/bare.py

@@ -0,0 +1,276 @@
+from __future__ import annotations
+import asyncio
+import datetime
+import json
+import os
+import re
+import sys
+import logging
+from typing import Callable
+from .repl import Agent, TabModel, CommandEngine, UIAdapter, HELP_TEXT
+from .gits import GitError, get_branch_base_sha, get_diff_between_refs, get_commit_history_with_stats, find_rev_commit, create_worktree, git_new_branch, git_new_branch_at
+logger = logging.getLogger(__name__)
+
+class BareAdapter:
+
+    def __init__(self, repl: 'BareRepl'):
+        self.repl = repl
+
+    def show_error(self, text: str) -> None:
+        print(f'\n✗ {text}', flush=True)
+
+    def show_command_result(self, cmd: str, content: str | object) -> None:
+        if isinstance(content, str):
+            print(content)
+        else:
+            print(str(content))
+
+    def show_diff(self, diff_text: str, ref1_label: str, ref2_label: str) -> None:
+        print(diff_text)
+
+    def show_history_list(self, lines: list[str]) -> None:
+        print('\n'.join(lines))
+
+    def show_history_raw(self, json_text: str) -> None:
+        print(json_text)
+
+    async def add_tab(self, tab: TabModel) -> None:
+        pass
+
+    def remove_tab(self, removed_index: int) -> None:
+        if self.repl.engine.active_index >= len(self.repl.engine.tabs):
+            self.repl.engine.active_index = len(self.repl.engine.tabs) - 1
+
+    def switch_to_tab(self, index: int) -> None:
+        self.repl._render_tab_delimiter()
+        self.repl._print_history_for_tab(self.repl.engine.tabs[index])
+
+    def refresh_chrome(self) -> None:
+        pass
+
+    def clear_tab_display(self, tab: TabModel) -> None:
+        pass
+
+    def on_agent_event(self, event: dict, tab: TabModel) -> None:
+        self.repl._handle_event(event, tab)
+
+    async def run_captured_shell(self, command: str, cwd: str, env: dict | None) -> str:
+        proc = await asyncio.create_subprocess_shell(command, cwd=cwd, stdout=asyncio.subprocess.PIPE, stderr=asyncio.subprocess.PIPE, env=env)
+        stdout, stderr = await proc.communicate()
+        out = stdout.decode(errors='replace').rstrip('\n')
+        err = stderr.decode(errors='replace').rstrip('\n')
+        body = out
+        if err:
+            body = (body + '\n' if body else '') + f'[stderr]\n{err}'
+        if proc.returncode:
+            body += f'\n[exit {proc.returncode}]'
+        return body
+
+    async def run_interactive_shell(self, command: str, cwd: str, env: dict | None) -> int:
+        proc = await asyncio.create_subprocess_shell(command, cwd=cwd, stdin=None, stdout=None, stderr=None, env=env)
+        await proc.wait()
+        return proc.returncode or 0
+
+    def exit_app(self, summary: list[dict]) -> None:
+        raise SystemExit
+
+class BareRepl:
+
+    def __init__(self, engine: CommandEngine, init_prompt: str | None=None):
+        self.engine = engine
+        self.adapter = BareAdapter(self)
+        self.engine.bind(self.adapter)
+        self.engine.attach_agent(self.engine.tabs[0])
+        self.init_prompt = init_prompt
+        self._pending_nls: int = 0
+        self._awaiting_content: bool = False
+        self._has_output: bool = False
+        self._last_stream_type: str | None = None
+
+    async def run(self) -> None:
+        loop = asyncio.get_running_loop()
+        if self.init_prompt:
+            p, self.init_prompt = (self.init_prompt, None)
+            await self.engine.active_tab.agent.run(p)
+        while True:
+            try:
+                line = await loop.run_in_executor(None, self._read_input)
+            except KeyboardInterrupt:
+                print('\n(Use /exit or Ctrl-D to quit)')
+                continue
+            except EOFError:
+                print()
+                break
+            if line is None:
+                break
+            line = line.strip()
+            if not line:
+                continue
+            if line.lower() in {'exit', 'quit'}:
+                break
+            await self.engine.handle_input(line)
+            tab = self.engine.active_tab
+            if tab.running_task is not None:
+                try:
+                    await tab.running_task
+                except asyncio.CancelledError:
+                    pass
+
+    def _read_input(self) -> str | None:
+        tab = self.engine.active_tab
+        prompt = f'[{tab.title}] ≫ '
+        lines: list[str] = []
+        while True:
+            try:
+                line = input(prompt)
+            except EOFError:
+                return None
+            lines.append(line)
+            if line.endswith('\\'):
+                lines[-1] = line[:-1]
+                prompt = '... '
+            else:
+                break
+        return '\n'.join(lines)
+
+    def _handle_event(self, event: dict, tab: TabModel) -> None:
+        t, p = (event['type'], event.get('payload', {}))
+        if t in ('text_delta', 'thinking_delta'):
+            delta = p.get('delta', '')
+            if delta:
+                self._write_delta(delta, t)
+        elif t == 'tool_start':
+            self._pending_nls = 0
+            self._awaiting_content = False
+            self._has_output = True
+            self._last_stream_type = t
+        elif t == 'tool_end':
+            result_msg = p.get('result', {})
+            content = result_msg.get('content')
+            is_err = p.get('is_error', False)
+            out_text = ''
+            if content:
+                parts: list[str] = []
+                if isinstance(content, str):
+                    parts.append(content)
+                elif isinstance(content, list):
+                    for block in content:
+                        if isinstance(block, dict) and block.get('type') == 'text':
+                            parts.append(block.get('text', ''))
+                out_text = '\n'.join(parts).strip('\n')
+            if is_err:
+                prefix = '✗ '
+                if not out_text:
+                    out_text = f'{p.get('name', '?')} failed'
+            else:
+                prefix = '→ ' if out_text else ''
+            if out_text:
+                self._write_delta(prefix + out_text, 'tool_result')
+            self._last_stream_type = t
+            print()
+        elif t == 'commit':
+            self._pending_nls = 0
+            self._awaiting_content = False
+            self._has_output = True
+            print(f'\n◇ [{p.get('sha', '')}] committed', flush=True)
+            self._last_stream_type = t
+        elif t == 'error':
+            self._pending_nls = 0
+            self._awaiting_content = False
+            self._has_output = True
+            err = str(p.get('error', p))
+            print(f'\n✗ {err}', flush=True)
+            self._last_stream_type = t
+        elif t in ('agent_start', 'turn_start'):
+            self._pending_nls = 0
+            self._awaiting_content = False
+            self._has_output = False
+            self._last_stream_type = None
+        elif t == 'agent_end':
+            self._pending_nls = 0
+            if self._has_output:
+                print()
+            self._last_stream_type = None
+            self._has_output = False
+            self._awaiting_content = False
+
+    def _write_delta(self, text: str, delta_type: str) -> None:
+        if delta_type != self._last_stream_type:
+            self._pending_nls = 0
+            self._awaiting_content = True
+            self._last_stream_type = delta_type
+        if self._awaiting_content:
+            text = text.lstrip('\n')
+            if not text:
+                return
+        if self._awaiting_content:
+            if self._has_output:
+                print()
+            self._awaiting_content = False
+        if not self._awaiting_content and self._pending_nls > 0:
+            print('\n' * self._pending_nls, end='', flush=True)
+            self._pending_nls = 0
+        rstripped = text.rstrip('\n')
+        if rstripped:
+            if delta_type == 'thinking_delta':
+                print(f'\x1b[2m{rstripped}\x1b[0m', end='', flush=True)
+            else:
+                print(rstripped, end='', flush=True)
+            self._has_output = True
+        self._pending_nls = len(text) - len(rstripped)
+
+    def _render_tab_delimiter(self) -> None:
+        tab_strs: list[str] = []
+        for i, t in enumerate(self.engine.tabs):
+            if i == self.engine.active_index:
+                tab_strs.append(f'\x1b[1m▶ {i + 1}. {t.title}\x1b[0m')
+            else:
+                tab_strs.append(f'\x1b[2m▷ {i + 1}. {t.title}\x1b[0m')
+        print('\n' + '┗━━┫ ' + ' ┃ '.join(tab_strs) + ' ┃')
+
+    def _print_history_for_tab(self, tab: TabModel) -> None:
+        for msg in tab.agent.messages:
+            role = msg.get('role', '')
+            content = msg.get('content', '')
+            is_error = msg.get('is_error', False)
+            if isinstance(content, list):
+                blocks = content
+            elif isinstance(content, str):
+                blocks = [{'type': 'text', 'text': content}]
+            else:
+                continue
+            if role == 'toolResult':
+                parts: list[str] = []
+                for block in blocks:
+                    if isinstance(block, dict) and block.get('type') == 'text':
+                        t = block.get('text', '').strip('\n')
+                        if t:
+                            parts.append(t)
+                if parts:
+                    prefix = '✗ ' if is_error else '→ '
+                    print(prefix + '\n'.join(parts))
+                continue
+            for block in blocks:
+                btype = block.get('type', 'text')
+                if btype == 'toolCall':
+                    args = block.get('arguments', {})
+                    if isinstance(args, dict):
+                        args = json.dumps(args, ensure_ascii=False)
+                    print(f'⚙ {block.get('name', '')} {args}')
+                    continue
+                text = block.get('text', '') or block.get('thinking', '') or ''
+                text = text.strip('\n')
+                if not text:
+                    continue
+                if btype == 'thinking':
+                    print(f'\x1b[2m{text}\x1b[0m')
+                elif is_error:
+                    print(f'✗ {text}')
+                elif role == 'user':
+                    print(f'≫ {text}')
+                elif role == 'commit':
+                    print(f'◇ {text}')
+                elif role == 'toolResult':
+                    print(f'→ {text}')
+                else:
+                    print(text)

{mlx_code-0.0.26 → mlx_code-0.0.28}/mlx_code/bats.py

@@ -1,3 +1,4 @@
+_UI_HTML = '<!DOCTYPE html>\n<html lang="en">\n<head>\n<meta charset="UTF-8">\n<meta name="viewport" content="width=device-width, initial-scale=1.0">\n<title>MLX Code</title>\n<style>\n*{margin:0;padding:0;box-sizing:border-box}\nbody{font-family:system-ui,-apple-system,sans-serif;background:#0d1117;color:#c9d1d9;height:100vh;display:flex;flex-direction:column}\n#hdr{padding:8px 16px;background:#161b22;border-bottom:1px solid #30363d;display:flex;justify-content:space-between;align-items:center;flex-shrink:0}\n#hdr h1{font-size:14px;font-weight:600}\n#hdr .info{display:flex;gap:8px;align-items:center;font-size:12px;color:#8b949e}\n#hdr button{background:transparent;color:#8b949e;border:1px solid #30363d;border-radius:6px;padding:4px 10px;cursor:pointer;font-size:12px}\n#hdr button:hover{color:#c9d1d9;border-color:#8b949e}\n#chat{flex:1;overflow-y:auto;padding:16px}\n.chat-inner{max-width:920px;margin:0 auto}\n.msg{margin-bottom:14px}\n.msg-role{font-size:12px;color:#8b949e;margin-bottom:3px}\n.msg-body{padding:10px 14px;border-radius:8px;line-height:1.6;white-space:pre-wrap;word-break:break-word;font-size:14px}\n.msg-user .msg-body{background:#1c2128;border:1px solid #30363d}\n.msg-assistant .msg-body{background:#161b22;border:1px solid #30363d}\n.msg-thinking .msg-body{color:#6e7681;font-style:italic;background:rgba(136,144,150,0.05);border-left:2px solid #30363d;font-size:13px}\n.msg-tool .msg-body{background:rgba(210,153,34,0.08);border-left:2px solid #d29922;font-family:ui-monospace,SFMono-Regular,monospace;font-size:13px}\n.msg-error .msg-body{color:#f85149}\n.cursor{display:inline-block;width:7px;height:15px;background:#58a6ff;animation:blink 1s steps(2) infinite;vertical-align:text-bottom;margin-left:2px;border-radius:1px}\n@keyframes blink{50%{opacity:0}}\n#input-area{padding:12px 16px;background:#161b22;border-top:1px solid #30363d;flex-shrink:0}\n.input-inner{max-width:920px;margin:0 auto;display:flex;gap:8px}\n#input{flex:1;background:#0d1117;color:#c9d1d9;border:1px solid #30363d;border-radius:8px;padding:10px 14px;font-family:inherit;font-size:14px;resize:none;height:44px;max-height:200px;line-height:1.5}\n#input:focus{outline:none;border-color:#58a6ff}\n#send{background:#238636;color:#fff;border:none;border-radius:8px;padding:0 20px;cursor:pointer;font-size:14px;font-weight:500;white-space:nowrap}\n#send:hover{background:#2ea043}\n#send:disabled{background:#21262d;color:#484f58;cursor:not-allowed}\n</style>\n</head>\n<body>\n<div id="hdr">\n  <h1>⚡ MLX Code</h1>\n  <div class="info">\n    <span id="status">connecting...</span>\n    <button onclick="clearChat()">Clear</button>\n  </div>\n</div>\n<div id="chat"><div class="chat-inner" id="chatInner"></div></div>\n<div id="input-area">\n  <div class="input-inner">\n    <textarea id="input" placeholder="Send a message... (Enter=send, Shift+Enter=newline)" rows="1"></textarea>\n    <button id="send" onclick="send()">Send</button>\n  </div>\n</div>\n<script>\nconst chatEl=document.getElementById(\'chat\');\nconst innerEl=document.getElementById(\'chatInner\');\nconst inputEl=document.getElementById(\'input\');\nconst sendBtn=document.getElementById(\'send\');\nconst statusEl=document.getElementById(\'status\');\nconst SYSTEM_PROMPT = \'You are a helpful assistant. You are running in a web chat mode with no tool execution capabilities. Answer the user directly and concisely.\';\nlet messages=[{role:\'system\',content:SYSTEM_PROMPT}];\nlet streaming=false;\n\ninputEl.addEventListener(\'keydown\',e=>{\n  if(e.key===\'Enter\'&&!e.shiftKey){e.preventDefault();send();}\n});\ninputEl.addEventListener(\'input\',()=>{\n  inputEl.style.height=\'auto\';\n  inputEl.style.height=Math.min(inputEl.scrollHeight,200)+\'px\';\n});\n\nfunction scrollBottom(){chatEl.scrollTop=chatEl.scrollHeight;}\n\nfunction addMsg(role,label){\n  const d=document.createElement(\'div\');\n  d.className=\'msg msg-\'+role;\n  const r=document.createElement(\'div\');r.className=\'msg-role\';r.textContent=label;\n  const b=document.createElement(\'div\');b.className=\'msg-body\';\n  d.appendChild(r);d.appendChild(b);\n  innerEl.appendChild(d);scrollBottom();return b;\n}\n\nfunction clearChat(){\n  if(streaming)return;\n  messages=[{role:\'system\',content:SYSTEM_PROMPT}];innerEl.innerHTML=\'\';inputEl.focus();\n}\n\nfunction stripToolXml(text){\n  // Remove complete <tool_call>...</tool_call> blocks\n  text=text.replace(/<tool_call>[\\s\\S]*?<\\/tool_call>/g,\'\');\n  // Handle incomplete <tool_call> at end (closing tag not yet received)\n  const idx=text.lastIndexOf(\'<tool_call>\');\n  if(idx!==-1&&text.indexOf(\'</tool_call>\',idx)===-1)return text.substring(0,idx);\n  // Handle partial opening tag at end (e.g. "<tool", "<tool_c")\n  const tag=\'<tool_call>\';\n  for(let i=tag.length-1;i>0;i--){\n    if(text.endsWith(tag.substring(0,i)))return text.substring(0,text.length-i);\n  }\n  return text;\n}\n\nfunction checkHealth(){\n  fetch(\'/health\').then(r=>r.json()).then(d=>{\n    statusEl.textContent=d.model||\'ready\';\n  }).catch(()=>{statusEl.textContent=\'offline\';});\n}\n\nasync function send(){\n  const text=inputEl.value.trim();\n  if(!text||streaming)return;\n  inputEl.value=\'\';inputEl.style.height=\'auto\';\n  messages.push({role:\'user\',content:text});\n  addMsg(\'user\',\'≫ You\').textContent=text;\n  streaming=true;sendBtn.disabled=true;statusEl.textContent=\'generating...\';\n\n  const aBody=addMsg(\'assistant\',\'○ Assistant\');\n  let tBody=null,displayText=\'\',thinkText=\'\',rawText=\'\',toolCalls=[];\n  const cursor=document.createElement(\'span\');cursor.className=\'cursor\';aBody.appendChild(cursor);\n\n  try{\n    const resp=await fetch(\'/v1/chat/completions\',{\n      method:\'POST\',\n      headers:{\'Content-Type\':\'application/json\'},\n      body:JSON.stringify({messages,max_tokens:8192})\n    });\n\n    if(!resp.ok){\n      cursor.remove();\n      let msg=\'HTTP \'+resp.status;\n      try{const e=await resp.json();msg+=\': \'+(e.error||JSON.stringify(e));}catch(_){try{msg+=\': \'+await resp.text();}catch(_){}}\n      aBody.textContent=\'✗ \'+msg;\n      aBody.parentElement.classList.add(\'msg-error\');\n      messages.pop();return;\n    }\n\n    const reader=resp.body.getReader();\n    const dec=new TextDecoder();\n    let buf=\'\';\n\n    while(true){\n      const{done,value}=await reader.read();\n      if(done)break;\n      buf+=dec.decode(value,{stream:true});\n      const lines=buf.split(\'\\n\');\n      buf=lines.pop();  // keep partial line in buffer\n\n      for(const line of lines){\n        if(!line.startsWith(\'data: \'))continue;\n        const data=line.slice(6).trim();\n        if(!data||data===\'[DONE]\')continue;\n        let ch;try{ch=JSON.parse(data);}catch(_){continue;}\n        const delta=ch.choices&&ch.choices[0]&&ch.choices[0].delta;\n        if(!delta)continue;\n\n        if(delta.reasoning_content){\n          if(!tBody){\n            cursor.remove();\n            tBody=addMsg(\'thinking\',\'◌ Thinking\');\n            tBody.appendChild(cursor.cloneNode());\n          }\n          thinkText+=delta.reasoning_content;\n          const c=tBody.querySelector(\'.cursor\');if(c)c.remove();\n          tBody.textContent=thinkText;\n          tBody.appendChild(cursor.cloneNode());\n          scrollBottom();\n        }\n\n        if(delta.content){\n          if(tBody){\n            const c=tBody.querySelector(\'.cursor\');if(c)c.remove();\n            tBody=null;aBody.appendChild(cursor);\n          }\n          rawText+=delta.content;\n          displayText=stripToolXml(rawText);\n          cursor.remove();aBody.textContent=displayText;aBody.appendChild(cursor);\n          scrollBottom();\n        }\n\n        if(delta.tool_calls){\n          cursor.remove();\n          for(const tc of delta.tool_calls){\n            const fn=tc.function||{};\n            if(fn.name){\n              toolCalls.push({name:fn.name,args:\'\'});\n              addMsg(\'tool\',\'⚙ \'+fn.name).textContent=fn.name;\n            }\n            if(fn.arguments&&toolCalls.length>0){\n              toolCalls[toolCalls.length-1].args+=fn.arguments;\n              const tbs=innerEl.querySelectorAll(\'.msg-tool .msg-body\');\n              if(tbs.length>0){\n                let disp=toolCalls[toolCalls.length-1].name+\'\\n\';\n                try{disp+=JSON.stringify(JSON.parse(toolCalls[toolCalls.length-1].args),null,2);}\n                catch(_){disp+=toolCalls[toolCalls.length-1].args;}\n                tbs[tbs.length-1].textContent=disp;\n              }\n            }\n          }\n          scrollBottom();\n        }\n      }\n    }\n\n    cursor.remove();\n    if(displayText.trim()){\n      aBody.textContent=displayText;\n      messages.push({role:\'assistant\',content:displayText});\n    }else{\n      aBody.textContent=thinkText?\'(thinking only)\':\'(no output)\';\n      messages.push({role:\'assistant\',content:displayText});\n    }\n    if(toolCalls.length>0){\n      addMsg(\'tool\',\'⚠ Note\').textContent=\'Tool calls cannot be executed in the web UI. Use the terminal REPL (--bare) for full tool support.\';\n    }\n  }catch(e){\n    cursor.remove();\n    aBody.textContent=\'✗ \'+e.message;\n    aBody.parentElement.classList.add(\'msg-error\');\n    if(messages.length>0&&messages[messages.length-1].role===\'user\')messages.pop();\n  }finally{\n    streaming=false;sendBtn.disabled=false;\n    checkHealth();inputEl.focus();\n  }\n}\n\ncheckHealth();inputEl.focus();\n</script>\n</body>\n</html>'
 import asyncio
 import json
 import queue as _queue
@@ -11,7 +12,7 @@ from pathlib import Path
 import mlx.core as mx
 from starlette.applications import Starlette
 from starlette.requests import Request
-from starlette.responses import StreamingResponse, JSONResponse
+from starlette.responses import StreamingResponse, JSONResponse, HTMLResponse
 from starlette.routing import Route
 import logging
 logger = logging.getLogger(__name__)

{mlx_code-0.0.26 → mlx_code-0.0.28}/mlx_code/main.py

@@ -944,6 +944,8 @@ def main():
     parser.add_argument('--skips', nargs='+', default=['(?m)^\\[SUGGESTION MODE[\\s\\S]*', '(?m)^<system-reminder>[\\s\\S]*?^</system-reminder>\\s*'], help='Regex patterns stripped from model output before it is returned to the client')
     parser.add_argument('--stream', default=None, help='File to stream log into')
     parser.add_argument('--bare', action='store_true', help='Use simple terminal REPL instead of TUI')
+    parser.add_argument('--web', action='store_true', help='Use web UI instead of TUI')
+    parser.add_argument('--web-port', type=int, default=None, help='Port for web UI (default: inference port + 80)')
     args, leash_args = parser.parse_known_args()
     logger.debug(f'args={args!r} leash_args={leash_args!r}')
     if args.engine == 'batch' and args.leash not in ('none', 'noapi'):
@@ -979,8 +981,13 @@ def main():
             if args.engine == 'cache':
                 threading.Thread(target=server.serve_forever, daemon=True).start()
             if args.leash == 'noapi':
-                from .repl import run_repl
-                run_repl(base_url=url, api=args.leash, repo=cwd, env=env, system=args.system, tool_names=args.tools, sdir=args.skill, init_prompt=args.prompt, resume=args.resume, stream=args.stream, bare=args.bare)
+                if args.web:
+                    from .web import run_web
+                    web_port = args.web_port if args.web_port is not None else port + 80
+                    run_web(base_url=url, api=args.leash, repo=cwd, env=env, system=args.system, tool_names=args.tools, sdir=args.skill, init_prompt=args.prompt, resume=args.resume, stream=args.stream, host=args.host, port=web_port)
+                else:
+                    from .repl import run_repl
+                    run_repl(base_url=url, api=args.leash, repo=cwd, env=env, system=args.system, tool_names=args.tools, sdir=args.skill, init_prompt=args.prompt, resume=args.resume, stream=args.stream, bare=args.bare)
             else:
                 env['GOOGLE_GEMINI_BASE_URL'] = url
                 env['GEMINI_API_KEY'] = 'mc'