zzz / openhands /agenthub /codeact_agent /function_calling.py
ar08's picture
Upload 1040 files
246d201 verified
"""This file contains the function calling implementation for different actions.
This is similar to the functionality of `CodeActResponseParser`.
"""
import json
from browsergym.core.action.highlevel import HighLevelActionSet
from litellm import (
ChatCompletionToolParam,
ChatCompletionToolParamFunctionChunk,
ModelResponse,
)
from openhands.core.exceptions import FunctionCallNotExistsError
from openhands.core.logger import openhands_logger as logger
from openhands.events.action import (
Action,
AgentDelegateAction,
AgentFinishAction,
BrowseInteractiveAction,
BrowseURLAction,
CmdRunAction,
FileEditAction,
FileReadAction,
IPythonRunCellAction,
MessageAction,
)
from openhands.events.event import FileEditSource, FileReadSource
from openhands.events.tool import ToolCallMetadata
_BASH_DESCRIPTION = """Execute a bash command in the terminal.
* Long running commands: For commands that may run indefinitely, it should be run in the background and the output should be redirected to a file, e.g. command = `python3 app.py > server.log 2>&1 &`.
* Interact with running process: If a bash command returns exit code `-1`, this means the process is not yet finished. By setting `is_input` to `true`, the assistant can interact with the running process and send empty `command` to retrieve any additional logs, or send additional text (set `command` to the text) to STDIN of the running process, or send command like `C-c` (Ctrl+C), `C-d` (Ctrl+D), `C-z` (Ctrl+Z) to interrupt the process.
"""
CmdRunTool = ChatCompletionToolParam(
type='function',
function=ChatCompletionToolParamFunctionChunk(
name='execute_bash',
description=_BASH_DESCRIPTION,
parameters={
'type': 'object',
'properties': {
'command': {
'type': 'string',
'description': 'The bash command to execute. Can be empty string to view additional logs when previous exit code is `-1`. Can be `C-c` (Ctrl+C) to interrupt the currently running process.',
},
'is_input': {
'type': 'string',
'description': 'If True, the command is an input to the running process. If False, the command is a bash command to be executed in the terminal. Default is False.',
'enum': ['true', 'false'],
},
},
'required': ['command'],
},
),
)
_IPYTHON_DESCRIPTION = """Run a cell of Python code in an IPython environment.
* The assistant should define variables and import packages before using them.
* The variable defined in the IPython environment will not be available outside the IPython environment (e.g., in terminal).
"""
IPythonTool = ChatCompletionToolParam(
type='function',
function=ChatCompletionToolParamFunctionChunk(
name='execute_ipython_cell',
description=_IPYTHON_DESCRIPTION,
parameters={
'type': 'object',
'properties': {
'code': {
'type': 'string',
'description': 'The Python code to execute. Supports magic commands like %pip.',
},
},
'required': ['code'],
},
),
)
_FILE_EDIT_DESCRIPTION = """Edit a file.
* The assistant can edit files by specifying the file path and providing a draft of the new file content.
* The draft content doesn't need to be exactly the same as the existing file; the assistant may skip unchanged lines using comments like `# unchanged` to indicate unchanged sections.
* IMPORTANT: For large files (e.g., > 300 lines), specify the range of lines to edit using `start` and `end` (1-indexed, inclusive). The range should be smaller than 300 lines.
* To append to a file, set both `start` and `end` to `-1`.
* If the file doesn't exist, a new file will be created with the provided content.
**Example 1: general edit for short files**
For example, given an existing file `/path/to/file.py` that looks like this:
(this is the end of the file)
1|class MyClass:
2| def __init__(self):
3| self.x = 1
4| self.y = 2
5| self.z = 3
6|
7|print(MyClass().z)
8|print(MyClass().x)
(this is the end of the file)
The assistant wants to edit the file to look like this:
(this is the end of the file)
1|class MyClass:
2| def __init__(self):
3| self.x = 1
4| self.y = 2
5|
6|print(MyClass().y)
(this is the end of the file)
The assistant may produce an edit action like this:
path="/path/to/file.txt" start=1 end=-1
content=```
class MyClass:
def __init__(self):
# no changes before
self.y = 2
# self.z is removed
# MyClass().z is removed
print(MyClass().y)
```
**Example 2: append to file for short files**
For example, given an existing file `/path/to/file.py` that looks like this:
(this is the end of the file)
1|class MyClass:
2| def __init__(self):
3| self.x = 1
4| self.y = 2
5| self.z = 3
6|
7|print(MyClass().z)
8|print(MyClass().x)
(this is the end of the file)
To append the following lines to the file:
```python
print(MyClass().y)
```
The assistant may produce an edit action like this:
path="/path/to/file.txt" start=-1 end=-1
content=```
print(MyClass().y)
```
**Example 3: edit for long files**
Given an existing file `/path/to/file.py` that looks like this:
(1000 more lines above)
1001|class MyClass:
1002| def __init__(self):
1003| self.x = 1
1004| self.y = 2
1005| self.z = 3
1006|
1007|print(MyClass().z)
1008|print(MyClass().x)
(2000 more lines below)
The assistant wants to edit the file to look like this:
(1000 more lines above)
1001|class MyClass:
1002| def __init__(self):
1003| self.x = 1
1004| self.y = 2
1005|
1006|print(MyClass().y)
(2000 more lines below)
The assistant may produce an edit action like this:
path="/path/to/file.txt" start=1001 end=1008
content=```
class MyClass:
def __init__(self):
# no changes before
self.y = 2
# self.z is removed
# MyClass().z is removed
print(MyClass().y)
```
"""
LLMBasedFileEditTool = ChatCompletionToolParam(
type='function',
function=ChatCompletionToolParamFunctionChunk(
name='edit_file',
description=_FILE_EDIT_DESCRIPTION,
parameters={
'type': 'object',
'properties': {
'path': {
'type': 'string',
'description': 'The absolute path to the file to be edited.',
},
'content': {
'type': 'string',
'description': 'A draft of the new content for the file being edited. Note that the assistant may skip unchanged lines.',
},
'start': {
'type': 'integer',
'description': 'The starting line number for the edit (1-indexed, inclusive). Default is 1.',
},
'end': {
'type': 'integer',
'description': 'The ending line number for the edit (1-indexed, inclusive). Default is -1 (end of file).',
},
},
'required': ['path', 'content'],
},
),
)
_STR_REPLACE_EDITOR_DESCRIPTION = """Custom editing tool for viewing, creating and editing files
* State is persistent across command calls and discussions with the user
* If `path` is a file, `view` displays the result of applying `cat -n`. If `path` is a directory, `view` lists non-hidden files and directories up to 2 levels deep
* The `create` command cannot be used if the specified `path` already exists as a file
* If a `command` generates a long output, it will be truncated and marked with `<response clipped>`
* The `undo_edit` command will revert the last edit made to the file at `path`
Notes for using the `str_replace` command:
* The `old_str` parameter should match EXACTLY one or more consecutive lines from the original file. Be mindful of whitespaces!
* If the `old_str` parameter is not unique in the file, the replacement will not be performed. Make sure to include enough context in `old_str` to make it unique
* The `new_str` parameter should contain the edited lines that should replace the `old_str`
"""
StrReplaceEditorTool = ChatCompletionToolParam(
type='function',
function=ChatCompletionToolParamFunctionChunk(
name='str_replace_editor',
description=_STR_REPLACE_EDITOR_DESCRIPTION,
parameters={
'type': 'object',
'properties': {
'command': {
'description': 'The commands to run. Allowed options are: `view`, `create`, `str_replace`, `insert`, `undo_edit`.',
'enum': ['view', 'create', 'str_replace', 'insert', 'undo_edit'],
'type': 'string',
},
'path': {
'description': 'Absolute path to file or directory, e.g. `/workspace/file.py` or `/workspace`.',
'type': 'string',
},
'file_text': {
'description': 'Required parameter of `create` command, with the content of the file to be created.',
'type': 'string',
},
'old_str': {
'description': 'Required parameter of `str_replace` command containing the string in `path` to replace.',
'type': 'string',
},
'new_str': {
'description': 'Optional parameter of `str_replace` command containing the new string (if not given, no string will be added). Required parameter of `insert` command containing the string to insert.',
'type': 'string',
},
'insert_line': {
'description': 'Required parameter of `insert` command. The `new_str` will be inserted AFTER the line `insert_line` of `path`.',
'type': 'integer',
},
'view_range': {
'description': 'Optional parameter of `view` command when `path` points to a file. If none is given, the full file is shown. If provided, the file will be shown in the indicated line number range, e.g. [11, 12] will show lines 11 and 12. Indexing at 1 to start. Setting `[start_line, -1]` shows all lines from `start_line` to the end of the file.',
'items': {'type': 'integer'},
'type': 'array',
},
},
'required': ['command', 'path'],
},
),
)
_WEB_DESCRIPTION = """Read (convert to markdown) content from a webpage. You should prefer using the `web_read` tool over the `browser` tool, but do use the `browser` tool if you need to interact with a webpage (e.g., click a button, fill out a form, etc.).
You may use the `web_read` tool to read content from a webpage, and even search the webpage content using a Google search query (e.g., url=`https://www.google.com/search?q=YOUR_QUERY`).
"""
WebReadTool = ChatCompletionToolParam(
type='function',
function=ChatCompletionToolParamFunctionChunk(
name='web_read',
description=_WEB_DESCRIPTION,
parameters={
'type': 'object',
'properties': {
'url': {
'type': 'string',
'description': 'The URL of the webpage to read. You can also use a Google search query here (e.g., `https://www.google.com/search?q=YOUR_QUERY`).',
}
},
'required': ['url'],
},
),
)
# from browsergym/core/action/highlevel.py
_browser_action_space = HighLevelActionSet(
subsets=['bid', 'nav'],
strict=False, # less strict on the parsing of the actions
multiaction=True, # enable to agent to take multiple actions at once
)
_BROWSER_DESCRIPTION = """Interact with the browser using Python code. Use it ONLY when you need to interact with a webpage.
See the description of "code" parameter for more details.
Multiple actions can be provided at once, but will be executed sequentially without any feedback from the page.
More than 2-3 actions usually leads to failure or unexpected behavior. Example:
fill('a12', 'example with "quotes"')
click('a51')
click('48', button='middle', modifiers=['Shift'])
"""
_BROWSER_TOOL_DESCRIPTION = """
The following 15 functions are available. Nothing else is supported.
goto(url: str)
Description: Navigate to a url.
Examples:
goto('http://www.example.com')
go_back()
Description: Navigate to the previous page in history.
Examples:
go_back()
go_forward()
Description: Navigate to the next page in history.
Examples:
go_forward()
noop(wait_ms: float = 1000)
Description: Do nothing, and optionally wait for the given time (in milliseconds).
You can use this to get the current page content and/or wait for the page to load.
Examples:
noop()
noop(500)
scroll(delta_x: float, delta_y: float)
Description: Scroll horizontally and vertically. Amounts in pixels, positive for right or down scrolling, negative for left or up scrolling. Dispatches a wheel event.
Examples:
scroll(0, 200)
scroll(-50.2, -100.5)
fill(bid: str, value: str)
Description: Fill out a form field. It focuses the element and triggers an input event with the entered text. It works for <input>, <textarea> and [contenteditable] elements.
Examples:
fill('237', 'example value')
fill('45', 'multi-line\nexample')
fill('a12', 'example with "quotes"')
select_option(bid: str, options: str | list[str])
Description: Select one or multiple options in a <select> element. You can specify option value or label to select. Multiple options can be selected.
Examples:
select_option('a48', 'blue')
select_option('c48', ['red', 'green', 'blue'])
click(bid: str, button: Literal['left', 'middle', 'right'] = 'left', modifiers: list[typing.Literal['Alt', 'Control', 'ControlOrMeta', 'Meta', 'Shift']] = [])
Description: Click an element.
Examples:
click('a51')
click('b22', button='right')
click('48', button='middle', modifiers=['Shift'])
dblclick(bid: str, button: Literal['left', 'middle', 'right'] = 'left', modifiers: list[typing.Literal['Alt', 'Control', 'ControlOrMeta', 'Meta', 'Shift']] = [])
Description: Double click an element.
Examples:
dblclick('12')
dblclick('ca42', button='right')
dblclick('178', button='middle', modifiers=['Shift'])
hover(bid: str)
Description: Hover over an element.
Examples:
hover('b8')
press(bid: str, key_comb: str)
Description: Focus the matching element and press a combination of keys. It accepts the logical key names that are emitted in the keyboardEvent.key property of the keyboard events: Backquote, Minus, Equal, Backslash, Backspace, Tab, Delete, Escape, ArrowDown, End, Enter, Home, Insert, PageDown, PageUp, ArrowRight, ArrowUp, F1 - F12, Digit0 - Digit9, KeyA - KeyZ, etc. You can alternatively specify a single character you'd like to produce such as "a" or "#". Following modification shortcuts are also supported: Shift, Control, Alt, Meta, ShiftLeft, ControlOrMeta. ControlOrMeta resolves to Control on Windows and Linux and to Meta on macOS.
Examples:
press('88', 'Backspace')
press('a26', 'ControlOrMeta+a')
press('a61', 'Meta+Shift+t')
focus(bid: str)
Description: Focus the matching element.
Examples:
focus('b455')
clear(bid: str)
Description: Clear the input field.
Examples:
clear('996')
drag_and_drop(from_bid: str, to_bid: str)
Description: Perform a drag & drop. Hover the element that will be dragged. Press left mouse button. Move mouse to the element that will receive the drop. Release left mouse button.
Examples:
drag_and_drop('56', '498')
upload_file(bid: str, file: str | list[str])
Description: Click an element and wait for a "filechooser" event, then select one or multiple input files for upload. Relative file paths are resolved relative to the current working directory. An empty list clears the selected files.
Examples:
upload_file('572', '/home/user/my_receipt.pdf')
upload_file('63', ['/home/bob/Documents/image.jpg', '/home/bob/Documents/file.zip'])
"""
for _, action in _browser_action_space.action_set.items():
assert (
action.signature in _BROWSER_TOOL_DESCRIPTION
), f'Browser description mismatch. Please double check if the BrowserGym updated their action space.\n\nAction: {action.signature}'
assert (
action.description in _BROWSER_TOOL_DESCRIPTION
), f'Browser description mismatch. Please double check if the BrowserGym updated their action space.\n\nAction: {action.description}'
BrowserTool = ChatCompletionToolParam(
type='function',
function=ChatCompletionToolParamFunctionChunk(
name='browser',
description=_BROWSER_DESCRIPTION,
parameters={
'type': 'object',
'properties': {
'code': {
'type': 'string',
'description': (
'The Python code that interacts with the browser.\n'
+ _BROWSER_TOOL_DESCRIPTION
),
}
},
'required': ['code'],
},
),
)
_FINISH_DESCRIPTION = """Finish the interaction when the task is complete OR if the assistant cannot proceed further with the task."""
FinishTool = ChatCompletionToolParam(
type='function',
function=ChatCompletionToolParamFunctionChunk(
name='finish',
description=_FINISH_DESCRIPTION,
),
)
def combine_thought(action: Action, thought: str) -> Action:
if not hasattr(action, 'thought'):
return action
if thought:
action.thought = thought
return action
def response_to_actions(response: ModelResponse) -> list[Action]:
actions: list[Action] = []
assert len(response.choices) == 1, 'Only one choice is supported for now'
assistant_msg = response.choices[0].message
if assistant_msg.tool_calls:
# Check if there's assistant_msg.content. If so, add it to the thought
thought = ''
if isinstance(assistant_msg.content, str):
thought = assistant_msg.content
elif isinstance(assistant_msg.content, list):
for msg in assistant_msg.content:
if msg['type'] == 'text':
thought += msg['text']
# Process each tool call to OpenHands action
for i, tool_call in enumerate(assistant_msg.tool_calls):
action: Action
try:
arguments = json.loads(tool_call.function.arguments)
except json.decoder.JSONDecodeError as e:
raise RuntimeError(
f'Failed to parse tool call arguments: {tool_call.function.arguments}'
) from e
if tool_call.function.name == 'execute_bash':
# this is an LLM error: add empty command to avoid breaking the tool call
if 'command' not in arguments:
arguments['command'] = ''
# convert is_input to boolean
if 'is_input' in arguments:
arguments['is_input'] = arguments['is_input'] == 'true'
action = CmdRunAction(**arguments)
elif tool_call.function.name == 'execute_ipython_cell':
action = IPythonRunCellAction(**arguments)
elif tool_call.function.name == 'delegate_to_browsing_agent':
action = AgentDelegateAction(
agent='BrowsingAgent',
inputs=arguments,
)
elif tool_call.function.name == 'finish':
action = AgentFinishAction()
elif tool_call.function.name == 'edit_file':
action = FileEditAction(**arguments)
elif tool_call.function.name == 'str_replace_editor':
# We implement this in agent_skills, which can be used via Jupyter
# convert tool_call.function.arguments to kwargs that can be passed to file_editor
code = f'print(file_editor(**{arguments}))'
logger.debug(
f'TOOL CALL: str_replace_editor -> file_editor with code: {code}'
)
if arguments['command'] == 'view':
action = FileReadAction(
path=arguments['path'],
translated_ipython_code=code,
impl_source=FileReadSource.OH_ACI,
)
else:
action = FileEditAction(
path=arguments['path'],
content='', # dummy value -- we don't need it
translated_ipython_code=code,
impl_source=FileEditSource.OH_ACI,
)
elif tool_call.function.name == 'browser':
action = BrowseInteractiveAction(browser_actions=arguments['code'])
elif tool_call.function.name == 'web_read':
action = BrowseURLAction(url=arguments['url'])
else:
raise FunctionCallNotExistsError(
f'Tool {tool_call.function.name} is not registered. (arguments: {arguments}). Please check the tool name and retry with an existing tool.'
)
# We only add thought to the first action
if i == 0:
action = combine_thought(action, thought)
# Add metadata for tool calling
action.tool_call_metadata = ToolCallMetadata(
tool_call_id=tool_call.id,
function_name=tool_call.function.name,
model_response=response,
total_calls_in_response=len(assistant_msg.tool_calls),
)
actions.append(action)
else:
actions.append(
MessageAction(content=assistant_msg.content, wait_for_response=True)
)
assert len(actions) >= 1
return actions
def get_tools(
codeact_enable_browsing: bool = False,
codeact_enable_llm_editor: bool = False,
codeact_enable_jupyter: bool = False,
) -> list[ChatCompletionToolParam]:
tools = [CmdRunTool, FinishTool]
if codeact_enable_browsing:
tools.append(WebReadTool)
tools.append(BrowserTool)
if codeact_enable_jupyter:
tools.append(IPythonTool)
if codeact_enable_llm_editor:
tools.append(LLMBasedFileEditTool)
else:
tools.append(StrReplaceEditorTool)
return tools