klaude-code 1.6.0__tar.gz → 1.7.0__tar.gz

{klaude_code-1.6.0 → klaude_code-1.7.0}/PKG-INFO

@@ -1,11 +1,12 @@
 Metadata-Version: 2.3
 Name: klaude-code
-Version: 1.6.0
+Version: 1.7.0
 Summary: Minimal code agent CLI
 Requires-Dist: anthropic>=0.66.0
 Requires-Dist: chardet>=5.2.0
 Requires-Dist: ddgs>=9.9.3
 Requires-Dist: diff-match-patch>=20241021
+Requires-Dist: google-genai>=1.56.0
 Requires-Dist: markdown-it-py>=4.0.0
 Requires-Dist: openai>=1.102.0
 Requires-Dist: pillow>=12.0.0
@@ -34,6 +35,10 @@ Minimal code agent CLI.
 - **Output truncation**: Large outputs saved to file system with snapshot links
 - **Skills**: Built-in + user + project Agent Skills (with implicit invocation by Skill tool or explicit invocation by typing `$`)
 - **Sessions**: Resumable with `--continue`
+- **Cost tracking**: Automatic API cost calculation and display (USD/CNY)
+- **Version update check**: Background PyPI version check with upgrade prompts
+- **Terminal title**: Shows current directory and model name
+- **Mermaid diagrams**: Interactive local HTML viewer with zoom, pan, and SVG export
 - **Extras**: Slash commands, sub-agents, image paste, terminal notifications, auto-theming
 
 ## Installation
@@ -77,6 +82,7 @@ klaude [--model <name>] [--select-model]
 - `--select-model`/`-s`: Open the interactive model selector at startup (shows all models unless `--model` is also provided).
 - `--continue`/`-c`: Resume the most recent session.
 - `--resume`/`-r`: Select a session to resume for this project.
+- `--resume-by-id <id>`: Resume a session by its ID directly.
 - `--vanilla`: Minimal mode with only basic tools (Bash, Read, Edit) and no system prompts.
 
 **Model selection behavior:**
@@ -251,12 +257,18 @@ klaude session clean-all
 
 Inside the interactive session (`klaude`), use these commands to streamline your workflow:
 
-- `/dev-doc [feature]` - Generate a comprehensive execution plan for a feature.
-- `/export` - Export last assistant message to a temp Markdown file.
-- `/init` - Bootstrap a new project structure or module.
 - `/model` - Switch the active LLM during the session.
+- `/thinking` - Configure model thinking/reasoning level.
 - `/clear` - Clear the current conversation context.
-- `/diff` - Show local git diff changes.
+- `/status` - Show session usage statistics (cost, tokens, model breakdown).
+- `/resume` - Select and resume a previous session.
+- `/fork-session` - Fork current session to a new session ID (supports interactive fork point selection).
+- `/export` - Export last assistant message to a temp Markdown file.
+- `/export-online` - Export and deploy session to surge.sh as a static webpage.
+- `/debug [filters]` - Toggle debug mode and configure debug filters.
+- `/init` - Bootstrap a new project structure or module.
+- `/dev-doc [feature]` - Generate a comprehensive execution plan for a feature.
+- `/terminal-setup` - Configure terminal for Shift+Enter support.
 - `/help` - List all available commands.
 
 
@@ -267,6 +279,8 @@ Inside the interactive session (`klaude`), use these commands to streamline your
 | `Enter`              | Submit input                                |
 | `Shift+Enter`        | Insert newline (requires `/terminal-setup`) |
 | `Ctrl+J`             | Insert newline                              |
+| `Ctrl+L`             | Open model picker overlay                   |
+| `Ctrl+T`             | Open thinking level picker overlay          |
 | `Ctrl+V`             | Paste image from clipboard                  |
 | `Left/Right`         | Move cursor (wraps across lines)            |
 | `Backspace`          | Delete character or selected text           |
@@ -290,4 +304,18 @@ echo "generate quicksort in python" | klaude exec --model gpt-5.1
 
 # Partial/ambiguous name opens the interactive selector (filtered)
 echo "generate quicksort in python" | klaude exec --model gpt
+
+# Stream all events as JSON lines (for programmatic processing)
+klaude exec "what is 2+2?" --stream-json
 ```
+
+### Sub-Agents
+
+The main agent can spawn specialized sub-agents for specific tasks:
+
+| Sub-Agent | Purpose |
+|-----------|---------|
+| **Explore** | Fast codebase exploration - find files, search code, answer questions about the codebase |
+| **Task** | Handle complex multi-step tasks autonomously |
+| **WebAgent** | Search the web, fetch pages, and analyze content |
+| **Oracle** | Advanced reasoning advisor for code reviews, architecture planning, and bug analysis |

{klaude_code-1.6.0 → klaude_code-1.7.0}/README.md

@@ -14,6 +14,10 @@ Minimal code agent CLI.
 - **Output truncation**: Large outputs saved to file system with snapshot links
 - **Skills**: Built-in + user + project Agent Skills (with implicit invocation by Skill tool or explicit invocation by typing `$`)
 - **Sessions**: Resumable with `--continue`
+- **Cost tracking**: Automatic API cost calculation and display (USD/CNY)
+- **Version update check**: Background PyPI version check with upgrade prompts
+- **Terminal title**: Shows current directory and model name
+- **Mermaid diagrams**: Interactive local HTML viewer with zoom, pan, and SVG export
 - **Extras**: Slash commands, sub-agents, image paste, terminal notifications, auto-theming
 
 ## Installation
@@ -57,6 +61,7 @@ klaude [--model <name>] [--select-model]
 - `--select-model`/`-s`: Open the interactive model selector at startup (shows all models unless `--model` is also provided).
 - `--continue`/`-c`: Resume the most recent session.
 - `--resume`/`-r`: Select a session to resume for this project.
+- `--resume-by-id <id>`: Resume a session by its ID directly.
 - `--vanilla`: Minimal mode with only basic tools (Bash, Read, Edit) and no system prompts.
 
 **Model selection behavior:**
@@ -231,12 +236,18 @@ klaude session clean-all
 
 Inside the interactive session (`klaude`), use these commands to streamline your workflow:
 
-- `/dev-doc [feature]` - Generate a comprehensive execution plan for a feature.
-- `/export` - Export last assistant message to a temp Markdown file.
-- `/init` - Bootstrap a new project structure or module.
 - `/model` - Switch the active LLM during the session.
+- `/thinking` - Configure model thinking/reasoning level.
 - `/clear` - Clear the current conversation context.
-- `/diff` - Show local git diff changes.
+- `/status` - Show session usage statistics (cost, tokens, model breakdown).
+- `/resume` - Select and resume a previous session.
+- `/fork-session` - Fork current session to a new session ID (supports interactive fork point selection).
+- `/export` - Export last assistant message to a temp Markdown file.
+- `/export-online` - Export and deploy session to surge.sh as a static webpage.
+- `/debug [filters]` - Toggle debug mode and configure debug filters.
+- `/init` - Bootstrap a new project structure or module.
+- `/dev-doc [feature]` - Generate a comprehensive execution plan for a feature.
+- `/terminal-setup` - Configure terminal for Shift+Enter support.
 - `/help` - List all available commands.
 
 
@@ -247,6 +258,8 @@ Inside the interactive session (`klaude`), use these commands to streamline your
 | `Enter`              | Submit input                                |
 | `Shift+Enter`        | Insert newline (requires `/terminal-setup`) |
 | `Ctrl+J`             | Insert newline                              |
+| `Ctrl+L`             | Open model picker overlay                   |
+| `Ctrl+T`             | Open thinking level picker overlay          |
 | `Ctrl+V`             | Paste image from clipboard                  |
 | `Left/Right`         | Move cursor (wraps across lines)            |
 | `Backspace`          | Delete character or selected text           |
@@ -270,4 +283,18 @@ echo "generate quicksort in python" | klaude exec --model gpt-5.1
 
 # Partial/ambiguous name opens the interactive selector (filtered)
 echo "generate quicksort in python" | klaude exec --model gpt
+
+# Stream all events as JSON lines (for programmatic processing)
+klaude exec "what is 2+2?" --stream-json
 ```
+
+### Sub-Agents
+
+The main agent can spawn specialized sub-agents for specific tasks:
+
+| Sub-Agent | Purpose |
+|-----------|---------|
+| **Explore** | Fast codebase exploration - find files, search code, answer questions about the codebase |
+| **Task** | Handle complex multi-step tasks autonomously |
+| **WebAgent** | Search the web, fetch pages, and analyze content |
+| **Oracle** | Advanced reasoning advisor for code reviews, architecture planning, and bug analysis |

{klaude_code-1.6.0 → klaude_code-1.7.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "uv_build"
 
 [project]
 name = "klaude-code"
-version = "1.6.0"
+version = "1.7.0"
 description = "Minimal code agent CLI"
 readme = "README.md"
 requires-python = ">=3.13"
@@ -13,6 +13,7 @@ dependencies = [
     "chardet>=5.2.0",
     "ddgs>=9.9.3",
     "diff-match-patch>=20241021",
+    "google-genai>=1.56.0",
     "markdown-it-py>=4.0.0",
     "openai>=1.102.0",
     "pillow>=12.0.0",

{klaude_code-1.6.0 → klaude_code-1.7.0}/src/klaude_code/cli/list_model.py

@@ -6,7 +6,7 @@ from rich.table import Table
 from rich.text import Text
 
 from klaude_code.config import Config
-from klaude_code.config.config import ModelConfig, ProviderConfig
+from klaude_code.config.config import ModelConfig, ProviderConfig, parse_env_var_syntax
 from klaude_code.protocol.llm_param import LLMClientProtocol
 from klaude_code.protocol.sub_agent import iter_sub_agent_profiles
 from klaude_code.ui.rich.theme import ThemeKey, get_theme
@@ -94,6 +94,29 @@ def format_api_key_display(provider: ProviderConfig) -> Text:
         return Text("N/A")
 
 
+def format_env_var_display(value: str | None) -> Text:
+    """Format environment variable display with warning if not set."""
+    env_var, resolved = parse_env_var_syntax(value)
+
+    if env_var:
+        # Using ${ENV_VAR} syntax
+        if resolved:
+            return Text.assemble(
+                (f"${{{env_var}}} = ", "dim"),
+                (mask_api_key(resolved), ""),
+            )
+        else:
+            return Text.assemble(
+                (f"${{{env_var}}} ", ""),
+                ("(not set)", ThemeKey.CONFIG_STATUS_ERROR),
+            )
+    elif value:
+        # Plain value
+        return Text(mask_api_key(value))
+    else:
+        return Text("N/A")
+
+
 def _get_model_params_display(model: ModelConfig) -> list[Text]:
     """Get display elements for model parameters."""
     params: list[Text] = []
@@ -162,15 +185,43 @@ def display_models_and_providers(config: Config):
                 format_api_key_display(provider),
             )
 
+        # AWS Bedrock parameters
+        if provider.protocol == LLMClientProtocol.BEDROCK:
+            if provider.aws_access_key:
+                provider_info.add_row(
+                    Text("AWS Key:", style=ThemeKey.CONFIG_PARAM_LABEL),
+                    format_env_var_display(provider.aws_access_key),
+                )
+            if provider.aws_secret_key:
+                provider_info.add_row(
+                    Text("AWS Secret:", style=ThemeKey.CONFIG_PARAM_LABEL),
+                    format_env_var_display(provider.aws_secret_key),
+                )
+            if provider.aws_region:
+                provider_info.add_row(
+                    Text("AWS Region:", style=ThemeKey.CONFIG_PARAM_LABEL),
+                    format_env_var_display(provider.aws_region),
+                )
+            if provider.aws_session_token:
+                provider_info.add_row(
+                    Text("AWS Token:", style=ThemeKey.CONFIG_PARAM_LABEL),
+                    format_env_var_display(provider.aws_session_token),
+                )
+            if provider.aws_profile:
+                provider_info.add_row(
+                    Text("AWS Profile:", style=ThemeKey.CONFIG_PARAM_LABEL),
+                    format_env_var_display(provider.aws_profile),
+                )
+
         # Check if provider has valid API key
         provider_available = not provider.is_api_key_missing()
 
         # Models table for this provider
         models_table = Table.grid(padding=(0, 1), expand=True)
         models_table.add_column(width=2, no_wrap=True)  # Status
-        models_table.add_column(overflow="fold", ratio=1)  # Name
-        models_table.add_column(overflow="fold", ratio=2)  # Model
-        models_table.add_column(overflow="fold", ratio=3)  # Params
+        models_table.add_column(overflow="fold", ratio=2)  # Name
+        models_table.add_column(overflow="fold", ratio=3)  # Model
+        models_table.add_column(overflow="fold", ratio=4)  # Params
 
         # Add header
         models_table.add_row(

{klaude_code-1.6.0 → klaude_code-1.7.0}/src/klaude_code/cli/session_cmd.py

@@ -22,8 +22,9 @@ def _session_confirm(sessions: list[Session.SessionMetaBrief], message: str) ->
     log(f"Sessions to delete ({len(sessions)}):")
     for s in sessions:
         msg_count_display = "N/A" if s.messages_count == -1 else str(s.messages_count)
-        first_msg = (s.first_user_message or "").strip().replace("\n", " ")[:50]
-        if len(s.first_user_message or "") > 50:
+        first_msg_text = s.user_messages[0] if s.user_messages else ""
+        first_msg = first_msg_text.strip().replace("\n", " ")[:50]
+        if len(first_msg_text) > 50:
             first_msg += "..."
         log(f"  {_fmt(s.updated_at)}  {msg_count_display:>3} msgs  {first_msg}")
 

{klaude_code-1.6.0 → klaude_code-1.7.0}/src/klaude_code/config/assets/builtin_config.yaml

@@ -7,7 +7,7 @@ provider_list:
   protocol: anthropic
   api_key: ${ANTHROPIC_API_KEY}
   model_list:
-  - model_name: sonnet
+  - model_name: sonnet@ant
     model_params:
       model: claude-sonnet-4-5-20250929
       context_limit: 200000
@@ -18,7 +18,7 @@ provider_list:
         output: 15.0
         cache_read: 0.3
         cache_write: 3.75
-  - model_name: opus
+  - model_name: opus@ant
     model_params:
       model: claude-opus-4-5-20251101
       context_limit: 200000
@@ -194,6 +194,41 @@ provider_list:
         output: 1.74
         cache_read: 0.04
 
+- provider_name: google
+  protocol: google
+  api_key: ${GOOGLE_API_KEY}
+  model_list:
+  - model_name: gemini-pro@google
+    model_params:
+      model: gemini-3-pro-preview
+      context_limit: 1048576
+      cost:
+        input: 2.0
+        output: 12.0
+        cache_read: 0.2
+  - model_name: gemini-flash@google
+    model_params:
+      model: gemini-3-flash-preview
+      context_limit: 1048576
+      cost:
+        input: 0.5
+        output: 3.0
+        cache_read: 0.05
+- provider_name: bedrock
+  protocol: bedrock
+  aws_access_key: ${AWS_ACCESS_KEY_ID}
+  aws_secret_key: ${AWS_SECRET_ACCESS_KEY}
+  aws_region: ${AWS_REGION}
+  model_list:
+  - model_name: sonnet@bedrock
+    model_params:
+      model: us.anthropic.claude-sonnet-4-5-20250929-v1:0
+      context_limit: 200000
+      cost:
+        input: 3.0
+        output: 15.0
+        cache_read: 0.3
+        cache_write: 3.75
 - provider_name: deepseek
   protocol: anthropic
   api_key: ${DEEPSEEK_API_KEY}

{klaude_code-1.6.0 → klaude_code-1.7.0}/src/klaude_code/config/builtin_config.py

@@ -17,6 +17,7 @@ if TYPE_CHECKING:
 # All supported API key environment variables
 SUPPORTED_API_KEY_ENVS = [
     "ANTHROPIC_API_KEY",
+    "GOOGLE_API_KEY",
     "OPENAI_API_KEY",
     "OPENROUTER_API_KEY",
     "DEEPSEEK_API_KEY",

{klaude_code-1.6.0 → klaude_code-1.7.0}/src/klaude_code/config/config.py

@@ -77,6 +77,7 @@ class ProviderConfig(llm_param.LLMConfigProviderParameter):
         """Check if the API key is missing (either not set or env var not found).
 
         For codex protocol, checks OAuth login status instead of API key.
+        For bedrock protocol, checks AWS credentials instead of API key.
         """
         from klaude_code.protocol.llm_param import LLMClientProtocol
 
@@ -89,6 +90,19 @@ class ProviderConfig(llm_param.LLMConfigProviderParameter):
             # Consider available if logged in and token not expired
             return state is None or state.is_expired()
 
+        if self.protocol == LLMClientProtocol.BEDROCK:
+            # Bedrock uses AWS credentials, not API key. Region is always required.
+            _, resolved_profile = parse_env_var_syntax(self.aws_profile)
+            _, resolved_region = parse_env_var_syntax(self.aws_region)
+
+            # When using profile, we still need region to initialize the client.
+            if resolved_profile:
+                return resolved_region is None
+
+            _, resolved_access_key = parse_env_var_syntax(self.aws_access_key)
+            _, resolved_secret_key = parse_env_var_syntax(self.aws_secret_key)
+            return resolved_region is None or resolved_access_key is None or resolved_secret_key is None
+
         return self.get_resolved_api_key() is None
 
 

{klaude_code-1.6.0 → klaude_code-1.7.0}/src/klaude_code/config/thinking.py

@@ -121,6 +121,13 @@ def format_current_thinking(config: llm_param.LLMConfigParameter) -> str:
             return f"enabled (budget_tokens={thinking.budget_tokens})"
         return "not set"
 
+    if protocol == llm_param.LLMClientProtocol.GOOGLE:
+        if thinking.type == "disabled":
+            return "off"
+        if thinking.type == "enabled":
+            return f"enabled (budget_tokens={thinking.budget_tokens})"
+        return "not set"
+
     return "unknown protocol"
 
 
@@ -230,6 +237,13 @@ def get_thinking_picker_data(config: llm_param.LLMConfigParameter) -> ThinkingPi
             current_value=_get_current_budget_value(thinking),
         )
 
+    if protocol == llm_param.LLMClientProtocol.GOOGLE:
+        return ThinkingPickerData(
+            options=_build_budget_options(),
+            message="Select thinking level:",
+            current_value=_get_current_budget_value(thinking),
+        )
+
     return None
 
 

klaude_code-1.7.0/src/klaude_code/llm/anthropic/client.py

@@ -0,0 +1,233 @@
+import json
+import os
+from collections.abc import AsyncGenerator
+from typing import Any, override
+
+import anthropic
+import httpx
+from anthropic import APIError
+from anthropic.types.beta.beta_input_json_delta import BetaInputJSONDelta
+from anthropic.types.beta.beta_raw_content_block_delta_event import BetaRawContentBlockDeltaEvent
+from anthropic.types.beta.beta_raw_content_block_start_event import BetaRawContentBlockStartEvent
+from anthropic.types.beta.beta_raw_content_block_stop_event import BetaRawContentBlockStopEvent
+from anthropic.types.beta.beta_raw_message_delta_event import BetaRawMessageDeltaEvent
+from anthropic.types.beta.beta_raw_message_start_event import BetaRawMessageStartEvent
+from anthropic.types.beta.beta_signature_delta import BetaSignatureDelta
+from anthropic.types.beta.beta_text_delta import BetaTextDelta
+from anthropic.types.beta.beta_thinking_delta import BetaThinkingDelta
+from anthropic.types.beta.beta_tool_use_block import BetaToolUseBlock
+from anthropic.types.beta.message_create_params import MessageCreateParamsStreaming
+
+from klaude_code import const
+from klaude_code.llm.anthropic.input import convert_history_to_input, convert_system_to_input, convert_tool_schema
+from klaude_code.llm.client import LLMClientABC
+from klaude_code.llm.input_common import apply_config_defaults
+from klaude_code.llm.registry import register
+from klaude_code.llm.usage import MetadataTracker
+from klaude_code.protocol import llm_param, model
+from klaude_code.trace import DebugType, log_debug
+
+
+def build_payload(param: llm_param.LLMCallParameter) -> MessageCreateParamsStreaming:
+    """Build Anthropic API request parameters."""
+    messages = convert_history_to_input(param.input, param.model)
+    tools = convert_tool_schema(param.tools)
+    system = convert_system_to_input(param.system)
+
+    payload: MessageCreateParamsStreaming = {
+        "model": str(param.model),
+        "tool_choice": {
+            "type": "auto",
+            "disable_parallel_tool_use": False,
+        },
+        "stream": True,
+        "max_tokens": param.max_tokens or const.DEFAULT_MAX_TOKENS,
+        "temperature": param.temperature or const.DEFAULT_TEMPERATURE,
+        "messages": messages,
+        "system": system,
+        "tools": tools,
+        "betas": ["interleaved-thinking-2025-05-14", "context-1m-2025-08-07"],
+    }
+
+    if param.thinking and param.thinking.type == "enabled":
+        payload["thinking"] = anthropic.types.ThinkingConfigEnabledParam(
+            type="enabled",
+            budget_tokens=param.thinking.budget_tokens or const.DEFAULT_ANTHROPIC_THINKING_BUDGET_TOKENS,
+        )
+
+    return payload
+
+
+async def parse_anthropic_stream(
+    stream: Any,
+    param: llm_param.LLMCallParameter,
+    metadata_tracker: MetadataTracker,
+) -> AsyncGenerator[model.ConversationItem]:
+    """Parse Anthropic beta messages stream and yield conversation items.
+
+    This function is shared between AnthropicClient and BedrockClient.
+    """
+    accumulated_thinking: list[str] = []
+    accumulated_content: list[str] = []
+    response_id: str | None = None
+
+    current_tool_name: str | None = None
+    current_tool_call_id: str | None = None
+    current_tool_inputs: list[str] | None = None
+
+    input_token = 0
+    cached_token = 0
+
+    async for event in await stream:
+        log_debug(
+            f"[{event.type}]",
+            event.model_dump_json(exclude_none=True),
+            style="blue",
+            debug_type=DebugType.LLM_STREAM,
+        )
+        match event:
+            case BetaRawMessageStartEvent() as event:
+                response_id = event.message.id
+                cached_token = event.message.usage.cache_read_input_tokens or 0
+                input_token = event.message.usage.input_tokens
+                yield model.StartItem(response_id=response_id)
+            case BetaRawContentBlockDeltaEvent() as event:
+                match event.delta:
+                    case BetaThinkingDelta() as delta:
+                        if delta.thinking:
+                            metadata_tracker.record_token()
+                        accumulated_thinking.append(delta.thinking)
+                        yield model.ReasoningTextDelta(
+                            content=delta.thinking,
+                            response_id=response_id,
+                        )
+                    case BetaSignatureDelta() as delta:
+                        yield model.ReasoningEncryptedItem(
+                            encrypted_content=delta.signature,
+                            response_id=response_id,
+                            model=str(param.model),
+                        )
+                    case BetaTextDelta() as delta:
+                        if delta.text:
+                            metadata_tracker.record_token()
+                        accumulated_content.append(delta.text)
+                        yield model.AssistantMessageDelta(
+                            content=delta.text,
+                            response_id=response_id,
+                        )
+                    case BetaInputJSONDelta() as delta:
+                        if current_tool_inputs is not None:
+                            if delta.partial_json:
+                                metadata_tracker.record_token()
+                            current_tool_inputs.append(delta.partial_json)
+                    case _:
+                        pass
+            case BetaRawContentBlockStartEvent() as event:
+                match event.content_block:
+                    case BetaToolUseBlock() as block:
+                        metadata_tracker.record_token()
+                        yield model.ToolCallStartItem(
+                            response_id=response_id,
+                            call_id=block.id,
+                            name=block.name,
+                        )
+                        current_tool_name = block.name
+                        current_tool_call_id = block.id
+                        current_tool_inputs = []
+                    case _:
+                        pass
+            case BetaRawContentBlockStopEvent():
+                if len(accumulated_thinking) > 0:
+                    metadata_tracker.record_token()
+                    full_thinking = "".join(accumulated_thinking)
+                    yield model.ReasoningTextItem(
+                        content=full_thinking,
+                        response_id=response_id,
+                        model=str(param.model),
+                    )
+                    accumulated_thinking.clear()
+                if len(accumulated_content) > 0:
+                    metadata_tracker.record_token()
+                    yield model.AssistantMessageItem(
+                        content="".join(accumulated_content),
+                        response_id=response_id,
+                    )
+                    accumulated_content.clear()
+                if current_tool_name and current_tool_call_id:
+                    metadata_tracker.record_token()
+                    yield model.ToolCallItem(
+                        name=current_tool_name,
+                        call_id=current_tool_call_id,
+                        arguments="".join(current_tool_inputs) if current_tool_inputs else "",
+                        response_id=response_id,
+                    )
+                    current_tool_name = None
+                    current_tool_call_id = None
+                    current_tool_inputs = None
+            case BetaRawMessageDeltaEvent() as event:
+                metadata_tracker.set_usage(
+                    model.Usage(
+                        input_tokens=input_token + cached_token,
+                        output_tokens=event.usage.output_tokens,
+                        cached_tokens=cached_token,
+                        context_size=input_token + cached_token + event.usage.output_tokens,
+                        context_limit=param.context_limit,
+                        max_tokens=param.max_tokens,
+                    )
+                )
+                metadata_tracker.set_model_name(str(param.model))
+                metadata_tracker.set_response_id(response_id)
+                yield metadata_tracker.finalize()
+            case _:
+                pass
+
+
+@register(llm_param.LLMClientProtocol.ANTHROPIC)
+class AnthropicClient(LLMClientABC):
+    def __init__(self, config: llm_param.LLMConfigParameter):
+        super().__init__(config)
+        # Remove ANTHROPIC_AUTH_TOKEN env var to prevent anthropic SDK from adding
+        # Authorization: Bearer header that may conflict with third-party APIs
+        # (e.g., deepseek, moonshot) that use Authorization header for authentication.
+        # The API key will be sent via X-Api-Key header instead.
+        saved_auth_token = os.environ.pop("ANTHROPIC_AUTH_TOKEN", None)
+        try:
+            client = anthropic.AsyncAnthropic(
+                api_key=config.api_key,
+                base_url=config.base_url,
+                timeout=httpx.Timeout(300.0, connect=15.0, read=285.0),
+            )
+        finally:
+            if saved_auth_token is not None:
+                os.environ["ANTHROPIC_AUTH_TOKEN"] = saved_auth_token
+        self.client: anthropic.AsyncAnthropic = client
+
+    @classmethod
+    @override
+    def create(cls, config: llm_param.LLMConfigParameter) -> "LLMClientABC":
+        return cls(config)
+
+    @override
+    async def call(self, param: llm_param.LLMCallParameter) -> AsyncGenerator[model.ConversationItem]:
+        param = apply_config_defaults(param, self.get_llm_config())
+
+        metadata_tracker = MetadataTracker(cost_config=self.get_llm_config().cost)
+
+        payload = build_payload(param)
+
+        log_debug(
+            json.dumps(payload, ensure_ascii=False, default=str),
+            style="yellow",
+            debug_type=DebugType.LLM_PAYLOAD,
+        )
+
+        stream = self.client.beta.messages.create(
+            **payload,
+            extra_headers={"extra": json.dumps({"session_id": param.session_id}, sort_keys=True)},
+        )
+
+        try:
+            async for item in parse_anthropic_stream(stream, param, metadata_tracker):
+                yield item
+        except (APIError, httpx.HTTPError) as e:
+            yield model.StreamErrorItem(error=f"{e.__class__.__name__} {e!s}")

klaude_code-1.7.0/src/klaude_code/llm/bedrock/__init__.py

@@ -0,0 +1,3 @@
+from klaude_code.llm.bedrock.client import BedrockClient
+
+__all__ = ["BedrockClient"]