| Name | agent-tool JSON |
| Version |
0.1.6
JSON |
| download |
| home_page | |
| Summary | A protocol for AI agent tools |
| upload_time | 2024-02-15 00:01:46 |
| maintainer | |
| docs_url | None |
| author | Patrick Barker |
| requires_python | >=3.10,<4.0 |
| license | Apache 2.0 |
| keywords |
|
| VCS |
|
| bugtrack_url |
|
| requirements |
No requirements were recorded.
|
| Travis-CI |
No Travis.
|
| coveralls test coverage |
No coveralls.
|
# Agent Tools
A common protocol for AI agent tools
## Installation
```
pip install agent-tool
```
## Usage
Let's define a simplified Selenium web browser tool
```python
from agent_tools import Tool, action, observation
from selenium import webdriver
class SeleniumBrowser(Tool):
"""Selenium browser as a tool"""
def __init__(self, headless: bool = True) -> None:
super().__init__()
options = webdriver.ChromeOptions()
if headless:
options.add_argument("--headless")
self.driver = webdriver.Chrome(options=options)
@action
def open_url(self, url: str) -> None:
"""Open a URL in the browser
Args:
url (str): URL to open
"""
self.driver.get(url)
@action
def click_element(self, selector: str, selector_type: str = "css_selector") -> None:
"""Click an element identified by a CSS selector
Args:
selector (str): CSS selector
selector_type (str, optional): Selector type. Defaults to "css_selector".
"""
element = self.driver.find_element(selector_type, selector)
element.click()
@observation
def get_html(self) -> str:
"""Get the entire HTML of the current page.
Returns:
str: Page HTML
"""
return self.driver.page_source
def close(self) -> None:
"""Close the tool"""
self.driver.quit()
```
We mark the functions to be made available to the agent as `@action` if they mutate the environment, and `@observation` if they are read only.
Now we can use this tool with an agent such as openai function calling
```python
from openai import OpenAI
client = OpenAI()
browser = SeleniumBrowser()
schemas = browser.json_schema()
messages = []
messages.append({"role": "system", "content": "Don't make assumptions about what values to plug into functions. Ask for clarification if a user request is ambiguous."})
messages.append({"role": "user", "content": "Get the HTML for the front page of wikipedia"})
headers = {
"Content-Type": "application/json",
"Authorization": "Bearer " + openai.api_key,
}
json_data = {"model": model, "messages": messages, "tools": schemas}
response = requests.post(
"https://api.openai.com/v1/chat/completions",
headers=headers,
json=json_data,
)
assistant_message = response.json()["choices"][0]["message"]
messages.append(assistant_message)
assistant_message
```
```json
{
"role": "assistant",
"tool_calls": [
{
"id": "call_RYXaDjxpUCfWmpXU7BZEYVqS",
"type": "function",
"function": {
"name": "open_url",
"arguments": "{\n \"url\": \"https://wikipedia.org\"}"
}
},
{
"id": "call_TJIWlknwdoinfWEMFNss",
"type": "function",
"function": {
"name": "get_html",
"arguments": ""
}
}
]
}
```
Then to use this action
```python
for tool in assistant_message["tool_calls"]:
action = browser.find_action(tool["function"]["name"])
args = json.loads(tool["function"]["arguments"])
resp = browser.use(action, **args)
```
Tools can be used locally or spun up on a server and used remotely (In progress)
## Share (In progress)
Register a tool with the AgentSea hub so others can find and use it.
```python
pip install agentsea
```
Create a repo to publish
```
agentsea create tool
```
Add your tool to the `tool.py` in the repo, fill in the `README.md` and add your dependencies using Poetry
Publish to the hub
```
agentsea publish .
```
## Roadmap
- Integrate with langchain, babyagi, autogpt, etc
Raw data
{
"_id": null,
"home_page": "",
"name": "agent-tool",
"maintainer": "",
"docs_url": null,
"requires_python": ">=3.10,<4.0",
"maintainer_email": "",
"keywords": "",
"author": "Patrick Barker",
"author_email": "patrickbarkerco@gmail.com",
"download_url": "https://files.pythonhosted.org/packages/be/0d/3d7bca58ae5c8b54462afcb74ff984bf5710785ad7d01fda275c413db7ed/agent_tool-0.1.6.tar.gz",
"platform": null,
"description": "# Agent Tools\n\nA common protocol for AI agent tools\n\n## Installation\n\n```\npip install agent-tool\n```\n\n## Usage\n\nLet's define a simplified Selenium web browser tool\n\n```python\nfrom agent_tools import Tool, action, observation\nfrom selenium import webdriver\n\n\nclass SeleniumBrowser(Tool):\n \"\"\"Selenium browser as a tool\"\"\"\n\n def __init__(self, headless: bool = True) -> None:\n super().__init__()\n options = webdriver.ChromeOptions()\n if headless:\n options.add_argument(\"--headless\")\n self.driver = webdriver.Chrome(options=options)\n\n @action\n def open_url(self, url: str) -> None:\n \"\"\"Open a URL in the browser\n\n Args:\n url (str): URL to open\n \"\"\"\n self.driver.get(url)\n\n @action\n def click_element(self, selector: str, selector_type: str = \"css_selector\") -> None:\n \"\"\"Click an element identified by a CSS selector\n\n Args:\n selector (str): CSS selector\n selector_type (str, optional): Selector type. Defaults to \"css_selector\".\n \"\"\"\n element = self.driver.find_element(selector_type, selector)\n element.click()\n\n @observation\n def get_html(self) -> str:\n \"\"\"Get the entire HTML of the current page.\n\n Returns:\n str: Page HTML\n \"\"\"\n return self.driver.page_source\n\n def close(self) -> None:\n \"\"\"Close the tool\"\"\"\n self.driver.quit()\n\n```\n\nWe mark the functions to be made available to the agent as `@action` if they mutate the environment, and `@observation` if they are read only.\n\nNow we can use this tool with an agent such as openai function calling\n\n```python\nfrom openai import OpenAI\n\nclient = OpenAI()\n\nbrowser = SeleniumBrowser()\nschemas = browser.json_schema()\n\nmessages = []\nmessages.append({\"role\": \"system\", \"content\": \"Don't make assumptions about what values to plug into functions. Ask for clarification if a user request is ambiguous.\"})\nmessages.append({\"role\": \"user\", \"content\": \"Get the HTML for the front page of wikipedia\"})\n\nheaders = {\n \"Content-Type\": \"application/json\",\n \"Authorization\": \"Bearer \" + openai.api_key,\n}\njson_data = {\"model\": model, \"messages\": messages, \"tools\": schemas}\nresponse = requests.post(\n \"https://api.openai.com/v1/chat/completions\",\n headers=headers,\n json=json_data,\n)\n\nassistant_message = response.json()[\"choices\"][0][\"message\"]\nmessages.append(assistant_message)\nassistant_message\n```\n\n```json\n{\n \"role\": \"assistant\",\n \"tool_calls\": [\n {\n \"id\": \"call_RYXaDjxpUCfWmpXU7BZEYVqS\",\n \"type\": \"function\",\n \"function\": {\n \"name\": \"open_url\",\n \"arguments\": \"{\\n \\\"url\\\": \\\"https://wikipedia.org\\\"}\"\n }\n },\n {\n \"id\": \"call_TJIWlknwdoinfWEMFNss\",\n \"type\": \"function\",\n \"function\": {\n \"name\": \"get_html\",\n \"arguments\": \"\"\n }\n }\n ]\n}\n```\n\nThen to use this action\n\n```python\nfor tool in assistant_message[\"tool_calls\"]:\n action = browser.find_action(tool[\"function\"][\"name\"])\n args = json.loads(tool[\"function\"][\"arguments\"])\n resp = browser.use(action, **args)\n```\n\nTools can be used locally or spun up on a server and used remotely (In progress)\n\n## Share (In progress)\n\nRegister a tool with the AgentSea hub so others can find and use it.\n\n```python\npip install agentsea\n```\n\nCreate a repo to publish\n\n```\nagentsea create tool\n```\n\nAdd your tool to the `tool.py` in the repo, fill in the `README.md` and add your dependencies using Poetry\n\nPublish to the hub\n\n```\nagentsea publish .\n```\n\n## Roadmap\n\n- Integrate with langchain, babyagi, autogpt, etc\n",
"bugtrack_url": null,
"license": "Apache 2.0",
"summary": "A protocol for AI agent tools",
"version": "0.1.6",
"project_urls": null,
"split_keywords": [],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "499d6039b3c0d2abf9316c4f52d6b261ef1b89f6227755dcbbc1077af7831768",
"md5": "ca15850b3278eda0ad66006ad3b437b6",
"sha256": "b0e3dedde2c4ddaa2a114b067315e24ea50a66c5f18123e96ffbac75e5128d26"
},
"downloads": -1,
"filename": "agent_tool-0.1.6-py3-none-any.whl",
"has_sig": false,
"md5_digest": "ca15850b3278eda0ad66006ad3b437b6",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.10,<4.0",
"size": 16914,
"upload_time": "2024-02-15T00:01:45",
"upload_time_iso_8601": "2024-02-15T00:01:45.533252Z",
"url": "https://files.pythonhosted.org/packages/49/9d/6039b3c0d2abf9316c4f52d6b261ef1b89f6227755dcbbc1077af7831768/agent_tool-0.1.6-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "be0d3d7bca58ae5c8b54462afcb74ff984bf5710785ad7d01fda275c413db7ed",
"md5": "b687be5c8c545748205d043c69000ed2",
"sha256": "4449d11a6a53462bc05848878b547f46f4d991c045e599211d7d7effb64c9f6b"
},
"downloads": -1,
"filename": "agent_tool-0.1.6.tar.gz",
"has_sig": false,
"md5_digest": "b687be5c8c545748205d043c69000ed2",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.10,<4.0",
"size": 11046,
"upload_time": "2024-02-15T00:01:46",
"upload_time_iso_8601": "2024-02-15T00:01:46.866731Z",
"url": "https://files.pythonhosted.org/packages/be/0d/3d7bca58ae5c8b54462afcb74ff984bf5710785ad7d01fda275c413db7ed/agent_tool-0.1.6.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2024-02-15 00:01:46",
"github": false,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"lcname": "agent-tool"
}
