# Chain of Thought Tool
A lightweight Python package that provides structured Chain of Thought reasoning capabilities for LLMs through function calling.
## Installation
```bash
pip install chain-of-thought-tool
```
Or install from source:
```bash
cd chain-of-thought-tool
pip install -e .
```
## Quick Start
```python
from chain_of_thought import TOOL_SPECS, HANDLERS
# Add to your LLM tools array
tools = [
*TOOL_SPECS, # Adds chain_of_thought_step, get_chain_summary, clear_chain
]
# In your tool handling logic
def handle_tool_call(tool_name, tool_args):
if tool_name in HANDLERS:
return HANDLERS[tool_name](**tool_args)
# ... handle other tools
```
## Usage with AWS Bedrock Converse API
```python
import boto3
from chain_of_thought import TOOL_SPECS, HANDLERS
bedrock = boto3.client('bedrock-runtime')
# Your conversation with tools
response = bedrock.converse(
modelId="anthropic.claude-3-5-sonnet-20241022-v2:0",
messages=[
{
"role": "user",
"content": [{"text": "Help me think through whether I should buy a house or keep renting."}]
}
],
toolConfig={
"tools": TOOL_SPECS # Just drop it in!
}
)
# Handle tool calls
for content in response['output']['message']['content']:
if content.get('toolUse'):
tool_use = content['toolUse']
tool_name = tool_use['name']
tool_args = tool_use['input']
# Execute the tool
result = HANDLERS[tool_name](**tool_args)
print(f"Tool {tool_name} result: {result}")
```
## Usage with OpenAI
```python
import openai
from chain_of_thought import TOOL_SPECS, HANDLERS
# Convert to OpenAI format
openai_tools = []
for tool in TOOL_SPECS:
openai_tools.append({
"type": "function",
"function": {
"name": tool["toolSpec"]["name"],
"description": tool["toolSpec"]["description"],
"parameters": tool["toolSpec"]["inputSchema"]["json"]
}
})
client = openai.OpenAI()
response = client.chat.completions.create(
model="gpt-4",
messages=[{"role": "user", "content": "Help me think through a complex decision."}],
tools=openai_tools
)
# Handle tool calls
if response.choices[0].message.tool_calls:
for tool_call in response.choices[0].message.tool_calls:
result = HANDLERS[tool_call.function.name](**eval(tool_call.function.arguments))
```
## How It Works
The Chain of Thought tool provides three main functions:
### 1. `chain_of_thought_step`
Process individual thoughts in a structured sequence with confidence tracking:
```python
{
"thought": "I need to consider the financial implications of buying vs renting",
"step_number": 1,
"total_steps": 5,
"next_step_needed": true,
"reasoning_stage": "Problem Definition",
"confidence": 0.8,
"evidence": ["Current market conditions", "Personal financial situation"],
"assumptions": ["Interest rates will remain stable"]
}
```
### 2. `get_chain_summary`
Get a comprehensive summary of the thinking process:
```python
# No arguments needed
{}
```
### 3. `clear_chain`
Reset the thinking process:
```python
# No arguments needed
{}
```
## Advanced Features
### Confidence Tracking
Each step can include a confidence level (0.0-1.0) to indicate certainty:
```python
{
"thought": "Based on my analysis, renting is more flexible",
"confidence": 0.85,
...
}
```
### Dependencies and Contradictions
Track relationships between thoughts:
```python
{
"thought": "This contradicts my earlier assumption",
"dependencies": [1, 2], # Depends on steps 1 and 2
"contradicts": [3], # Contradicts step 3
...
}
```
### Evidence and Assumptions
Make reasoning transparent:
```python
{
"evidence": ["Market data shows 5% annual appreciation"],
"assumptions": ["My job will remain stable"],
...
}
```
### Structured Stages
Guide thinking through defined stages:
- `Problem Definition`
- `Research`
- `Analysis`
- `Synthesis`
- `Conclusion`
## Why This Approach?
**Traditional Problems:**
- ❌ MCP tools require separate server processes
- ❌ Framework-specific tools (LangChain, etc.)
- ❌ Complex infrastructure for simple functions
**Our Solution:**
- ✅ Simple `pip install` and import
- ✅ Works with any LLM API (OpenAI, Anthropic, etc.)
- ✅ Self-contained tool specs and implementations
- ✅ Zero infrastructure - just Python functions
- ✅ Structured reasoning with confidence tracking
## Thread Safety
For production use with multiple concurrent conversations:
```python
from chain_of_thought import ThreadAwareChainOfThought
# Create isolated instance per conversation
cot = ThreadAwareChainOfThought(conversation_id="user-123")
tools = cot.get_tool_specs()
handlers = cot.get_handlers()
# Use in your conversation
response = bedrock.converse(
toolConfig={"tools": tools},
# ...
)
# Handle with thread-specific handlers
result = handlers[tool_name](**tool_args)
```
## Contributing
This project demonstrates pluggable LLM tools. Contributions welcome for:
- Improved reasoning capabilities
- Additional metadata tracking
- Better summarization algorithms
- Integration helpers for more platforms
## License
MIT License
Raw data
{
"_id": null,
"home_page": null,
"name": "chain-of-thought-tool",
"maintainer": null,
"docs_url": null,
"requires_python": ">=3.8",
"maintainer_email": null,
"keywords": "llm, function-calling, reasoning, ai, tools, chain-of-thought, cot",
"author": "Code Developer",
"author_email": "Code Developer <code-developer@democratize.technology>",
"download_url": "https://files.pythonhosted.org/packages/36/11/8ee425f81b4740961a1f803368fe56f45e9c91af1de8abdd5b2e9d123a0d/chain_of_thought_tool-0.1.0.tar.gz",
"platform": null,
"description": "# Chain of Thought Tool\n\nA lightweight Python package that provides structured Chain of Thought reasoning capabilities for LLMs through function calling.\n\n## Installation\n\n```bash\npip install chain-of-thought-tool\n```\n\nOr install from source:\n```bash\ncd chain-of-thought-tool\npip install -e .\n```\n\n## Quick Start\n\n```python\nfrom chain_of_thought import TOOL_SPECS, HANDLERS\n\n# Add to your LLM tools array\ntools = [\n *TOOL_SPECS, # Adds chain_of_thought_step, get_chain_summary, clear_chain\n]\n\n# In your tool handling logic\ndef handle_tool_call(tool_name, tool_args):\n if tool_name in HANDLERS:\n return HANDLERS[tool_name](**tool_args)\n # ... handle other tools\n```\n\n## Usage with AWS Bedrock Converse API\n\n```python\nimport boto3\nfrom chain_of_thought import TOOL_SPECS, HANDLERS\n\nbedrock = boto3.client('bedrock-runtime')\n\n# Your conversation with tools\nresponse = bedrock.converse(\n modelId=\"anthropic.claude-3-5-sonnet-20241022-v2:0\",\n messages=[\n {\n \"role\": \"user\", \n \"content\": [{\"text\": \"Help me think through whether I should buy a house or keep renting.\"}]\n }\n ],\n toolConfig={\n \"tools\": TOOL_SPECS # Just drop it in!\n }\n)\n\n# Handle tool calls\nfor content in response['output']['message']['content']:\n if content.get('toolUse'):\n tool_use = content['toolUse']\n tool_name = tool_use['name']\n tool_args = tool_use['input']\n \n # Execute the tool\n result = HANDLERS[tool_name](**tool_args)\n print(f\"Tool {tool_name} result: {result}\")\n```\n\n## Usage with OpenAI\n\n```python\nimport openai\nfrom chain_of_thought import TOOL_SPECS, HANDLERS\n\n# Convert to OpenAI format\nopenai_tools = []\nfor tool in TOOL_SPECS:\n openai_tools.append({\n \"type\": \"function\",\n \"function\": {\n \"name\": tool[\"toolSpec\"][\"name\"],\n \"description\": tool[\"toolSpec\"][\"description\"],\n \"parameters\": tool[\"toolSpec\"][\"inputSchema\"][\"json\"]\n }\n })\n\nclient = openai.OpenAI()\nresponse = client.chat.completions.create(\n model=\"gpt-4\",\n messages=[{\"role\": \"user\", \"content\": \"Help me think through a complex decision.\"}],\n tools=openai_tools\n)\n\n# Handle tool calls\nif response.choices[0].message.tool_calls:\n for tool_call in response.choices[0].message.tool_calls:\n result = HANDLERS[tool_call.function.name](**eval(tool_call.function.arguments))\n```\n\n## How It Works\n\nThe Chain of Thought tool provides three main functions:\n\n### 1. `chain_of_thought_step`\nProcess individual thoughts in a structured sequence with confidence tracking:\n\n```python\n{\n \"thought\": \"I need to consider the financial implications of buying vs renting\",\n \"step_number\": 1,\n \"total_steps\": 5,\n \"next_step_needed\": true,\n \"reasoning_stage\": \"Problem Definition\",\n \"confidence\": 0.8,\n \"evidence\": [\"Current market conditions\", \"Personal financial situation\"],\n \"assumptions\": [\"Interest rates will remain stable\"]\n}\n```\n\n### 2. `get_chain_summary`\nGet a comprehensive summary of the thinking process:\n\n```python\n# No arguments needed\n{}\n```\n\n### 3. `clear_chain`\nReset the thinking process:\n\n```python\n# No arguments needed \n{}\n```\n\n## Advanced Features\n\n### Confidence Tracking\nEach step can include a confidence level (0.0-1.0) to indicate certainty:\n\n```python\n{\n \"thought\": \"Based on my analysis, renting is more flexible\",\n \"confidence\": 0.85,\n ...\n}\n```\n\n### Dependencies and Contradictions\nTrack relationships between thoughts:\n\n```python\n{\n \"thought\": \"This contradicts my earlier assumption\",\n \"dependencies\": [1, 2], # Depends on steps 1 and 2\n \"contradicts\": [3], # Contradicts step 3\n ...\n}\n```\n\n### Evidence and Assumptions\nMake reasoning transparent:\n\n```python\n{\n \"evidence\": [\"Market data shows 5% annual appreciation\"],\n \"assumptions\": [\"My job will remain stable\"],\n ...\n}\n```\n\n### Structured Stages\nGuide thinking through defined stages:\n- `Problem Definition`\n- `Research` \n- `Analysis`\n- `Synthesis`\n- `Conclusion`\n\n## Why This Approach?\n\n**Traditional Problems:**\n- \u274c MCP tools require separate server processes\n- \u274c Framework-specific tools (LangChain, etc.)\n- \u274c Complex infrastructure for simple functions\n\n**Our Solution:**\n- \u2705 Simple `pip install` and import\n- \u2705 Works with any LLM API (OpenAI, Anthropic, etc.)\n- \u2705 Self-contained tool specs and implementations\n- \u2705 Zero infrastructure - just Python functions\n- \u2705 Structured reasoning with confidence tracking\n\n## Thread Safety\n\nFor production use with multiple concurrent conversations:\n\n```python\nfrom chain_of_thought import ThreadAwareChainOfThought\n\n# Create isolated instance per conversation\ncot = ThreadAwareChainOfThought(conversation_id=\"user-123\")\ntools = cot.get_tool_specs()\nhandlers = cot.get_handlers()\n\n# Use in your conversation\nresponse = bedrock.converse(\n toolConfig={\"tools\": tools},\n # ...\n)\n\n# Handle with thread-specific handlers\nresult = handlers[tool_name](**tool_args)\n```\n\n## Contributing\n\nThis project demonstrates pluggable LLM tools. Contributions welcome for:\n- Improved reasoning capabilities\n- Additional metadata tracking\n- Better summarization algorithms\n- Integration helpers for more platforms\n\n## License\n\nMIT License\n",
"bugtrack_url": null,
"license": null,
"summary": "A lightweight Chain of Thought reasoning tool for LLM function calling",
"version": "0.1.0",
"project_urls": {
"Changelog": "https://github.com/democratize-technology/chain-of-thought-tool/releases",
"Homepage": "https://github.com/democratize-technology/chain-of-thought-tool",
"Issues": "https://github.com/democratize-technology/chain-of-thought-tool/issues",
"Repository": "https://github.com/democratize-technology/chain-of-thought-tool"
},
"split_keywords": [
"llm",
" function-calling",
" reasoning",
" ai",
" tools",
" chain-of-thought",
" cot"
],
"urls": [
{
"comment_text": null,
"digests": {
"blake2b_256": "c522cb5b44db9f93ca3b34812bb0a831c6d72db8eb4ffca65f53329aef562bb9",
"md5": "abf470c93c7ed3647c58f986e186c5d1",
"sha256": "35248ef23fd7ee57d62f60f8e7537028264216dead008c21bd96aea59d2ab03f"
},
"downloads": -1,
"filename": "chain_of_thought_tool-0.1.0-py3-none-any.whl",
"has_sig": false,
"md5_digest": "abf470c93c7ed3647c58f986e186c5d1",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.8",
"size": 16251,
"upload_time": "2025-09-03T02:48:07",
"upload_time_iso_8601": "2025-09-03T02:48:07.066889Z",
"url": "https://files.pythonhosted.org/packages/c5/22/cb5b44db9f93ca3b34812bb0a831c6d72db8eb4ffca65f53329aef562bb9/chain_of_thought_tool-0.1.0-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": null,
"digests": {
"blake2b_256": "36118ee425f81b4740961a1f803368fe56f45e9c91af1de8abdd5b2e9d123a0d",
"md5": "fae3ed3f61655653d40c08b617311b1a",
"sha256": "4b497d57cf3d24c070abbfa02f78a8f91f926860cf913a1cd5d18788996cd58d"
},
"downloads": -1,
"filename": "chain_of_thought_tool-0.1.0.tar.gz",
"has_sig": false,
"md5_digest": "fae3ed3f61655653d40c08b617311b1a",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.8",
"size": 27031,
"upload_time": "2025-09-03T02:48:08",
"upload_time_iso_8601": "2025-09-03T02:48:08.466185Z",
"url": "https://files.pythonhosted.org/packages/36/11/8ee425f81b4740961a1f803368fe56f45e9c91af1de8abdd5b2e9d123a0d/chain_of_thought_tool-0.1.0.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2025-09-03 02:48:08",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "democratize-technology",
"github_project": "chain-of-thought-tool",
"travis_ci": false,
"coveralls": false,
"github_actions": false,
"lcname": "chain-of-thought-tool"
}
JSON
