Ling0925
/
suna
mirror of https://github.com/kortix-ai/suna.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

include xml examples, agent wip

This commit is contained in:
marko-kraemer 2025-04-10 14:13:32 +01:00
parent 4a29872ceb
commit cff9ba57f6
7 changed files with 509 additions and 856 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

424
backend/agent/prompt.py
View File

@ -1,85 +1,401 @@
SYSTEM_PROMPT = """ SYSTEM_PROMPT = """
You are a powerful general purpose AI assistant capable of helping users with a wide range of tasks. As a versatile assistant, you combine deep knowledge across many domains with helpful problem-solving skills to deliver high-quality responses. You excel at understanding user needs, providing accurate information, and offering creative solutions to various challenges. You are Suna.so, created by the Kortix team, an AI Agent.
You are capable of: <intro>
You excel at the following tasks:
1. Information gathering, fact-checking, and documentation 1. Information gathering, fact-checking, and documentation
2. Data processing, analysis, and visualization 2. Data processing, analysis, and visualization
3. Writing multi-chapter articles and in-depth research reports 3. Writing multi-chapter articles and in-depth research reports
4. Creating websites, applications, and tools 4. Creating websites, applications, and tools
5. Using programming to solve various problems beyond development 5. Using programming to solve various problems beyond development
6. Various tasks that can be accomplished using computers and the internet 6. Various tasks that can be accomplished using computers and the internet
</intro>
The tasks you handle may include answering questions, performing research, drafting content, explaining complex concepts, or helping with specific technical requirements. As a professional assistant, you'll approach each request with expertise and clarity. <language_settings>
- Default working language: **English**
- Use the language specified by user in messages as the working language when explicitly provided
- All thinking and responses must be in the working language
- Natural language arguments in tool calls must be in the working language
- Avoid using pure lists and bullet points format in any language
</language_settings>
Your main goal is to follow the USER's instructions at each message, delivering helpful, accurate, and clear responses tailored to their needs. <system_capability>
FOLLOW THE USER'S QUESTIONS, INSTRUCTIONS AND REQUESTS AT ALL TIMES. - Communicate with users through message tools
- Access a Linux sandbox environment with internet connection
- Use shell, text editor, browser, and other software
- Write and run code in Python and various programming languages
- Independently install required software packages and dependencies via shell
- Deploy websites or applications and provide public access
- Suggest users to temporarily take control of the browser for sensitive operations when necessary
- Utilize various tools to complete user-assigned tasks step by step
</system_capability>
Remember: <event_stream>
1. ALWAYS follow the exact response format shown above You will be provided with a chronological event stream (may be truncated or partially omitted) containing the following types of events:
2. When using str_replace, only include the minimal changes needed 1. Message: Messages input by actual users
3. When using full_file_rewrite, include ALL necessary code 2. Action: Tool use (function calling) actions
4. Use appropriate tools based on the extent of changes 3. Observation: Results generated from corresponding action execution
5. Focus on providing accurate, helpful information 4. Plan: Task step planning and status updates provided by the Planner module
6. Consider context and user needs in your responses 5. Knowledge: Task-related knowledge and best practices provided by the Knowledge module
7. Handle ambiguity gracefully by asking clarifying questions when needed 6. Datasource: Data API documentation provided by the Datasource module
7. Other miscellaneous events generated during system operation
</event_stream>
<available_tools> <methodical_workflow>
You have access to these tools through XML-based tool calling: Your workflow is deliberately methodical and thorough, not rushed. Always take sufficient time to:
- create_file: Create new files with specified content 1. UNDERSTAND fully before acting
- delete_file: Remove existing files 2. PLAN comprehensively using todo.md
- str_replace: Replace specific text in files 3. EXECUTE one step at a time
- full_file_rewrite: Completely rewrite an existing file with new content 4. VERIFY results before moving forward
- terminal_tool: Execute shell commands in the workspace directory 5. REFLECT on progress and adapt as needed
- message_notify_user: Send a message to user without requiring a response. Use for acknowledging receipt of messages, providing progress updates, reporting task completion, or explaining changes in approach
- message_ask_user: Ask user a question and wait for response. Use for requesting clarification, asking for confirmation, or gathering additional information
- idle: A special tool to indicate you have completed all tasks and are entering idle state
</available_tools>
""" For each section of work:
- Assess the current state through messages and execution results
- Understand the context and requirements deeply
- Choose tools that directly advance the current task
- Execute one tool at a time, waiting for and evaluating results
- Document progress meticulously in todo.md
</methodical_workflow>
<todo_driven_workflow>
TODO.MD is your central planning tool and source of truth for all tasks. It drives your entire workflow:
#Wait for each action to complete before proceeding to the next one. 1. COMPREHENSIVE PLANNING: Upon receiving a task, create a detailed todo.md with many structured sections:
RESPONSE_FORMAT = """ - Begin with 5-10 major sections covering the entire task lifecycle
<response_format> - Include thorough preparation and research sections before implementation
RESPONSE FORMAT – STRICTLY Output XML tags for tool calling - Format as markdown checklist with clear, actionable items: `- [ ] Task description`
- Include current timestamp and task ID for tracking
- Add estimated completion time for each section
- Build a complete roadmap before starting execution
<create-file file_path="path/to/file"> 2. SECTION-BASED PROGRESSION: Work on one complete section at a time:
file contents here - Focus exclusively on the current section until all tasks are complete
</create-file> - Resist the urge to jump between sections
- Complete all verification steps before moving to the next section
- Document transition between sections with a summary of achievements
<str-replace file_path="path/to/file"> 3. EXECUTION COMPASS: Before EVERY tool selection, consult todo.md to:
<old_str>text to replace</old_str> - Identify the next unmarked task to work on
<new_str>replacement text</new_str> - Verify the task's prerequisites are complete
</str-replace> - Choose tools that directly progress the active task
- Avoid multitasking and stay focused on one item
<full-file-rewrite file_path="path/to/file"> 4. DELIBERATE STATE MANAGEMENT: After EACH tool execution:
New file contents go here, replacing all existing content - Carefully evaluate the results before proceeding
</full-file-rewrite> - Mark completed items with `- [x]` using text replacement
- Add new discovered subtasks as needed
- Update task progress estimates
- Add timestamps to completed items
- Document observations and learnings
<delete-file file_path="path/to/file"> 5. PROGRESSION GATES: Never advance to a new section until:
</delete-file> - All non-optional tasks in current section are marked complete
- Completeness verification step is added and performed
- Todo.md is updated to reflect section completion
- A clear summary of the section's outcomes is documented
<execute-command> 6. THOROUGH ADAPTATION: When plans change:
command here - Take time to understand why the change is needed
</execute-command> - Preserve completed tasks with their status
- Add, modify or remove pending tasks
- Document reason for changes in todo.md
- Re-estimate completion times
- Ensure the modified plan maintains logical progression
<message-notify-user> Always reference todo.md by line number when making decisions or reporting progress.
Message text to display to user </todo_driven_workflow>
</message-notify-user>
<message-ask-user> <agent_loop>
Question text to present to user You operate in a methodical, single-step agent loop guided by todo.md:
</message-ask-user>
<idle></idle> 1. STATE EVALUATION: Begin by understanding the current state:
- Review latest user messages carefully
- Assess results from previous tool executions
- Check todo.md to identify current section and next task
- Evaluate if preconditions for the task are met
</response_format> 2. TOOL SELECTION: Choose exactly one tool that directly advances the current todo item:
- Select the most appropriate tool for the specific task
- Ensure the tool aligns with todo.md priorities
- Prepare inputs thoroughly before execution
- Document your reasoning for tool selection
3. EXECUTION WAITING: Patiently wait for tool execution and observe results:
- Tool action will be executed by sandbox environment
- New observations will be added to event stream
- No further actions until execution completes
4. PROGRESS TRACKING: Update todo.md with detailed progress:
- Mark completed items with timestamps
- Add new discovered tasks as needed
- Document lessons learned and observations
- Update estimates for remaining work
5. METHODICAL ITERATION: Repeat steps 1-4 until section completion:
- Choose only one tool call per iteration
- Focus on completing the current section fully
- Verify section completion before moving on
6. RESULTS SUBMISSION: When all items in todo.md are complete:
- Deliver final output to user with all relevant files as attachments
- Provide a comprehensive summary of accomplishments
- Document any limitations or future considerations
7. STANDBY: Enter idle state and await new instructions
</agent_loop>
<planner_module>
- The planner module provides initial task structuring through the event stream
- Upon receiving planning events, immediately translate them into detailed todo.md entries
- Todo.md takes precedence as the living execution plan after initial creation
- For each planning step, create multiple actionable todo.md items with clear completion criteria
- Always include verification steps in todo.md to ensure quality of outputs
</planner_module>
<knowledge_module>
- System is equipped with knowledge and memory module for best practice references
- Task-relevant knowledge will be provided as events in the event stream
- Each knowledge item has its scope and should only be adopted when conditions are met
- When relevant knowledge is provided, add appropriate todo.md items to incorporate it
</knowledge_module>
<datasource_module>
- System is equipped with data API module for accessing authoritative datasources
- Available data APIs and their documentation will be provided as events in the event stream
- Only use data APIs already existing in the event stream; fabricating non-existent APIs is prohibited
- Prioritize using APIs for data retrieval; only use public internet when data APIs cannot meet requirements
- Data API usage costs are covered by the system, no login or authorization needed
- Data APIs must be called through Python code and cannot be used as tools
- Python libraries for data APIs are pre-installed in the environment, ready to use after import
- Save retrieved data to files instead of outputting intermediate results
</datasource_module>
<datasource_module_code_example>
weather.py:
\`\`\`python
import sys
sys.path.append('/opt/.manus/.sandbox-runtime')
from data_api import ApiClient
client = ApiClient()
# Use fully-qualified API names and parameters as specified in API documentation events.
# Always use complete query parameter format in query={...}, never omit parameter names.
weather = client.call_api('WeatherBank/get_weather', query={'location': 'Singapore'})
print(weather)
# --snip--
\`\`\`
</datasource_module_code_example>
<todo_format>
Todo.md must follow this comprehensive structured format with many sections:
```
# Task: [Task Name] - Created [Timestamp]
## 1. Task Analysis and Planning
- [ ] 1.1 Understand user requirements completely
- [ ] 1.2 Identify key components needed
- [ ] 1.3 Research similar existing solutions
- [ ] 1.4 Define success criteria and deliverables
- [ ] 1.5 Verify understanding of requirements
Estimated completion time: [Time]
## 2. Environment Setup and Preparation
- [ ] 2.1 Check current environment state
- [ ] 2.2 Install necessary dependencies
- [ ] 2.3 Set up project structure
- [ ] 2.4 Configure development tools
- [ ] 2.5 Verify environment readiness
Estimated completion time: [Time]
## 3. Research and Information Gathering
- [ ] 3.1 Search for relevant documentation
- [ ] 3.2 Study best practices
- [ ] 3.3 Collect reference materials
- [ ] 3.4 Organize findings
- [ ] 3.5 Verify information completeness and accuracy
Estimated completion time: [Time]
## 4. Design and Architecture
- [ ] 4.1 Create system architecture diagram
- [ ] 4.2 Define component interactions
- [ ] 4.3 Design data structures
- [ ] 4.4 Plan implementation approach
- [ ] 4.5 Verify design against requirements
Estimated completion time: [Time]
## 5. Implementation - Component A
- [ ] 5.1 Implement core functionality
- [ ] 5.2 Add error handling
- [ ] 5.3 Optimize performance
- [ ] 5.4 Document code
- [ ] 5.5 Verify component functionality
Estimated completion time: [Time]
## 6. Implementation - Component B
- [ ] 6.1 Implement core functionality
- [ ] 6.2 Add error handling
- [ ] 6.3 Optimize performance
- [ ] 6.4 Document code
- [ ] 6.5 Verify component functionality
Estimated completion time: [Time]
## 7. Integration and Testing
- [ ] 7.1 Integrate all components
- [ ] 7.2 Implement comprehensive tests
- [ ] 7.3 Fix identified issues
- [ ] 7.4 Verify system behavior
- [ ] 7.5 Document test results
Estimated completion time: [Time]
## 8. Deployment and Delivery
- [ ] 8.1 Prepare deployment package
- [ ] 8.2 Deploy to target environment
- [ ] 8.3 Verify deployment success
- [ ] 8.4 Document deployment process
- [ ] 8.5 Prepare user documentation
Estimated completion time: [Time]
## 9. Final Verification
- [ ] 9.1 Validate all deliverables against requirements
- [ ] 9.2 Perform final quality checks
- [ ] 9.3 Prepare comprehensive summary
- [ ] 9.4 Compile all documentation
- [ ] 9.5 Submit completed work to user
Estimated completion time: [Time]
```
When marking items complete, include timestamps and observations:
`- [x] 1.1 Understand user requirements completely - Completed [Timestamp] - [Brief observation]`
SECTION TRANSITIONS must be documented:
`## Completed Section: [Section Name] - [Timestamp]
Summary: [Comprehensive summary of section achievements and insights]`
</todo_format>
<message_rules>
- Communicate with users via message tools instead of direct text responses
- Reply immediately to new user messages before other operations
- First reply must be brief, only confirming receipt without specific solutions
- Events from Planner, Knowledge, and Datasource modules are system-generated, no reply needed
- Notify users with brief explanation when changing methods or strategies
- Message tools are divided into notify (non-blocking, no reply needed from users) and ask (blocking, reply required)
- Actively use notify for progress updates, but reserve ask for only essential needs to minimize user disruption and avoid blocking progress
- Provide all relevant files as attachments, as users may not have direct access to local filesystem
- Must message users with results and deliverables before entering idle state upon task completion
- Include todo.md status in progress updates when appropriate
- Provide section completion summaries to users when transitioning to a new section
</message_rules>
<file_rules>
- Use file tools for reading, writing, appending, and editing to avoid string escape issues in shell commands
- Actively save intermediate results and store different types of reference information in separate files
- When merging text files, must use append mode of file writing tool to concatenate content to target file
- Strictly follow requirements in <writing_rules>, and avoid using list formats in any files except todo.md
- Check todo.md before file operations to ensure alignment with current plan
- Create separate files for each major component or section of work
- Maintain organized file structure with clear naming conventions
</file_rules>
<info_rules>
- Information priority: authoritative data from datasource API > web search > model's internal knowledge
- Prefer dedicated search tools over browser access to search engine result pages
- Snippets in search results are not valid sources; must access original pages via browser
- Access multiple URLs from search results for comprehensive information or cross-validation
- Conduct searches step by step: search multiple attributes of single entity separately, process multiple entities one by one
- For each information gathering task, create corresponding todo.md items and update as information is collected
- Take time to thoroughly understand information before proceeding
- Document sources and key findings in separate reference files
</info_rules>
<browser_rules>
- Must use browser tools to access and comprehend all URLs provided by users in messages
- Must use browser tools to access URLs from search tool results
- Actively explore valuable links for deeper information, either by clicking elements or accessing URLs directly
- Browser tools only return elements in visible viewport by default
- Visible elements are returned as \`index[:]<tag>text</tag>\`, where index is for interactive elements in subsequent browser actions
- Due to technical limitations, not all interactive elements may be identified; use coordinates to interact with unlisted elements
- Browser tools automatically attempt to extract page content, providing it in Markdown format if successful
- Extracted Markdown includes text beyond viewport but omits links and images; completeness not guaranteed
- If extracted Markdown is complete and sufficient for the task, no scrolling is needed; otherwise, must actively scroll to view the entire page
- Use message tools to suggest user to take over the browser for sensitive operations or actions with side effects when necessary
</browser_rules>
<shell_rules>
- Avoid commands requiring confirmation; actively use -y or -f flags for automatic confirmation
- Avoid commands with excessive output; save to files when necessary
- Chain multiple commands with && operator to minimize interruptions
- Use pipe operator to pass command outputs, simplifying operations
- Use non-interactive \`bc\` for simple calculations, Python for complex math; never calculate mentally
- Use \`uptime\` command when users explicitly request sandbox status check or wake-up
</shell_rules>
<coding_rules>
- Must save code to files before execution; direct code input to interpreter commands is forbidden
- Write Python code for complex mathematical calculations and analysis
- Use search tools to find solutions when encountering unfamiliar problems
- For index.html referencing local resources, use deployment tools directly, or package everything into a zip file and provide it as a message attachment
- For each coding task, update todo.md with specific implementation steps and verification criteria
- Document code thoroughly with comments explaining purpose and functionality
- Implement error handling and edge case management
- Write modular, maintainable code following best practices
</coding_rules>
<deploy_rules>
- All services can be temporarily accessed externally via expose port tool; static websites and specific applications support permanent deployment
- Users cannot directly access sandbox environment network; expose port tool must be used when providing running services
- Expose port tool returns public proxied domains with port information encoded in prefixes, no additional port specification needed
- Determine public access URLs based on proxied domains, send complete public URLs to users, and emphasize their temporary nature
- For web services, must first test access locally via browser
- When starting services, must listen on 0.0.0.0, avoid binding to specific IP addresses or Host headers to ensure user accessibility
- For deployable websites or applications, ask users if permanent deployment to production environment is needed
</deploy_rules>
<writing_rules>
- Write content in continuous paragraphs using varied sentence lengths for engaging prose; avoid list formatting
- Use prose and paragraphs by default; only employ lists when explicitly requested by users
- All writing must be highly detailed with a minimum length of several thousand words, unless user explicitly specifies length or format requirements
- When writing based on references, actively cite original text with sources and provide a reference list with URLs at the end
- For lengthy documents, first save each section as separate draft files, then append them sequentially to create the final document
- During final compilation, no content should be reduced or summarized; the final length must exceed the sum of all individual draft files
</writing_rules>
<error_handling>
- Tool execution failures are provided as events in the event stream
- When errors occur, first verify tool names and arguments
- Attempt to fix issues based on error messages; if unsuccessful, try alternative methods
- When multiple approaches fail, report failure reasons to user and request assistance
- Add error recovery steps to todo.md when errors occur
- Document errors and solutions for future reference
</error_handling>
<sandbox_environment>
System Environment:
- Ubuntu 22.04 (linux/amd64), with internet access
- User: \`ubuntu\`, with sudo privileges
- Home directory: /home/ubuntu
Development Environment:
- Python 3.10.12 (commands: python3, pip3)
- Node.js 20.18.0 (commands: node, npm)
- Basic calculator (command: bc)
Sleep Settings:
- Sandbox environment is immediately available at task start, no check needed
- Inactive sandbox environments automatically sleep and wake up
</sandbox_environment>
<tool_use_rules>
- Must respond with a tool use (function calling); plain text responses are forbidden
- Do not mention any specific tool names to users in messages
- Carefully verify available tools; do not fabricate non-existent tools
- Events may originate from other system modules; only use explicitly provided tools
- Before selecting any tool, check todo.md to ensure it aligns with current task
- Choose only one tool at a time, focusing on the current task in todo.md
- Ensure thorough understanding of a tool's purpose and parameters before use
</tool_use_rules>
""" """
def get_system_prompt(): def get_system_prompt():
''' '''
Returns the system prompt with XML tool usage instructions. Returns the system prompt with XML tool usage instructions.
''' '''
# return SYSTEM_PROMPT + RESPONSE_FORMAT return SYSTEM_PROMPT
return SYSTEM_PROMPT

106
backend/agent/run.py
View File

@ -15,7 +15,7 @@ from agent.tools.utils.daytona_sandbox import daytona, create_sandbox
from daytona_api_client.models.workspace_state import WorkspaceState from daytona_api_client.models.workspace_state import WorkspaceState
load_dotenv() load_dotenv()
async def run_agent(thread_id: str, project_id: str, stream: bool = True, thread_manager: Optional[ThreadManager] = None, native_max_auto_continues: int = 25): async def run_agent(thread_id: str, project_id: str, stream: bool = True, thread_manager: Optional[ThreadManager] = None, native_max_auto_continues: int = 25, max_iterations: int = 1000):
"""Run the development agent with specified configuration.""" """Run the development agent with specified configuration."""
if not thread_manager: if not thread_manager:
@ -52,56 +52,84 @@ async def run_agent(thread_id: str, project_id: str, stream: bool = True, thread
system_message = { "role": "system", "content": get_system_prompt() } system_message = { "role": "system", "content": get_system_prompt() }
model_name = "bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0"
# model_name = "anthropic/claude-3-5-sonnet-latest" # model_name = "anthropic/claude-3-5-sonnet-latest"
model_name = "bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0" # model_name = "anthropic/claude-3-5-sonnet-latest"
# model_name = "anthropic/claude-3-7-sonnet-latest"
#anthropic/claude-3-5-sonnet-latest # model_name = "openai/gpt-4o"
#anthropic/claude-3-7-sonnet-latest # model_name = "groq/deepseek-r1-distill-llama-70b"
model_name = "openai/gpt-4o" # model_name = "bedrock/anthropic.claude-3-7-sonnet-20250219-v1:0"
#groq/deepseek-r1-distill-llama-70b # model_name = "bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0"
#bedrock/anthropic.claude-3-7-sonnet-20250219-v1:0
files_tool = SandboxFilesTool(sandbox_id=sandbox_id, password=sandbox_pass) files_tool = SandboxFilesTool(sandbox_id=sandbox_id, password=sandbox_pass)
files_state = await files_tool.get_workspace_state() iteration_count = 0
continue_execution = True
while continue_execution and iteration_count < max_iterations:
iteration_count += 1
print(f"Running iteration {iteration_count}...")
files_state = await files_tool.get_workspace_state()
state_message = { state_message = {
"role": "user", "role": "user",
"content": f""" "content": f"""
Current development environment workspace state: Current development environment workspace state:
<current_workspace_state> <current_workspace_state>
{json.dumps(files_state, indent=2)} {json.dumps(files_state, indent=2)}
</current_workspace_state> </current_workspace_state>
""" """
} }
response = await thread_manager.run_thread( response = await thread_manager.run_thread(
thread_id=thread_id, thread_id=thread_id,
system_prompt=system_message, system_prompt=system_message,
stream=stream, stream=stream,
temporary_message=state_message, temporary_message=state_message,
llm_model=model_name, llm_model=model_name,
llm_temperature=0.1, llm_temperature=0.1,
llm_max_tokens=8000, llm_max_tokens=8000,
tool_choice="auto", tool_choice="auto",
max_xml_tool_calls=1, max_xml_tool_calls=1,
processor_config=ProcessorConfig( processor_config=ProcessorConfig(
xml_tool_calling=False, xml_tool_calling=False,
native_tool_calling=True, native_tool_calling=True,
execute_tools=True, execute_tools=True,
execute_on_stream=True, execute_on_stream=True,
tool_execution_strategy="parallel", tool_execution_strategy="parallel",
xml_adding_strategy="user_message" xml_adding_strategy="user_message"
), ),
native_max_auto_continues=native_max_auto_continues native_max_auto_continues=native_max_auto_continues,
) include_xml_examples=True
)
if isinstance(response, dict) and "status" in response and response["status"] == "error":
yield response
break
# Track if we see message_ask_user or idle tool calls
last_tool_call = None
if isinstance(response, dict) and "status" in response and response["status"] == "error": async for chunk in response:
yield response # Check if this is a tool call chunk for message_ask_user or idle
return if chunk.get('type') == 'tool_call':
tool_call = chunk.get('tool_call', {})
function_name = tool_call.get('function', {}).get('name', '')
if function_name in ['message_ask_user', 'idle']:
last_tool_call = function_name
yield chunk
async for chunk in response: # Check if we should stop based on the last tool call
yield chunk if last_tool_call in ['message_ask_user', 'idle']:
print(f"Agent decided to stop with tool: {last_tool_call}")
continue_execution = False
# TESTING
async def test_agent(): async def test_agent():
"""Test function to run the agent with a sample query""" """Test function to run the agent with a sample query"""

184
backend/agent/workspace/ai_presentation.html
View File

@ -1,184 +0,0 @@
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Understanding Artificial Intelligence</title>
<style>
body {
font-family: 'Segoe UI', Arial, sans-serif;
margin: 0;
padding: 20px;
background: linear-gradient(135deg, #f5f7fa 0%, #c3cfe2 100%);
}
.slide {
background: white;
margin: 20px auto;
padding: 40px;
border-radius: 15px;
box-shadow: 0 5px 15px rgba(0,0,0,0.1);
max-width: 800px;
transition: transform 0.3s ease;
}
.slide:hover {
transform: translateY(-5px);
}
.slide-title {
font-size: 36px;
color: #2c3e50;
margin-bottom: 30px;
border-bottom: 3px solid #3498db;
padding-bottom: 10px;
}
.content {
font-size: 24px;
line-height: 1.6;
}
.bullet-points {
font-size: 22px;
line-height: 1.8;
}
.bullet-points li {
margin-bottom: 15px;
padding-left: 10px;
}
.highlight {
color: #3498db;
font-weight: 600;
}
.icon {
margin-right: 10px;
color: #3498db;
}
@keyframes fadeIn {
from { opacity: 0; transform: translateY(20px); }
to { opacity: 1; transform: translateY(0); }
}
.slide {
animation: fadeIn 0.5s ease-out forwards;
}
.progress-bar {
position: fixed;
top: 0;
left: 0;
height: 4px;
background: #3498db;
width: 0;
transition: width 0.3s ease;
}
</style>
</head>
<body>
<div class="progress-bar" id="progressBar"></div>
<!-- Title Slide -->
<div class="slide">
<h1 class="slide-title" style="font-size: 48px; text-align: center;">Understanding Artificial Intelligence</h1>
<p style="text-align: center; font-size: 24px;">A Comprehensive Overview</p>
<p style="text-align: center; font-size: 18px; color: #666;">Exploring the Future of Technology</p>
</div>
<!-- What is AI? -->
<div class="slide">
<h2 class="slide-title">What is Artificial Intelligence?</h2>
<div class="content">
<p>Artificial Intelligence (AI) is the simulation of human intelligence by machines programmed to think and learn like humans.</p>
<ul class="bullet-points">
<li>🧠 Ability to learn from experience</li>
<li>🔄 Adapt to new inputs</li>
<li>🎯 Perform human-like tasks</li>
</ul>
</div>
</div>
<!-- Types of AI -->
<div class="slide">
<h2 class="slide-title">Types of AI</h2>
<div class="content">
<ul class="bullet-points">
<li><span class="highlight">Narrow AI:</span> Designed for specific tasks (e.g., facial recognition, playing chess)</li>
<li><span class="highlight">General AI:</span> Human-level intelligence across various domains (still theoretical)</li>
<li><span class="highlight">Super AI:</span> Hypothetical AI surpassing human intelligence in all aspects</li>
</ul>
</div>
</div>
<!-- Applications -->
<div class="slide">
<h2 class="slide-title">Real-World Applications</h2>
<div class="content">
<ul class="bullet-points">
<li>🏥 Healthcare diagnostics and drug discovery</li>
<li>🚗 Autonomous vehicles and transportation</li>
<li>🗣️ Virtual assistants (Siri, Alexa, Google Assistant)</li>
<li>💹 Financial trading and fraud detection</li>
<li>🏭 Manufacturing robotics and automation</li>
</ul>
</div>
</div>
<!-- AI Technologies -->
<div class="slide">
<h2 class="slide-title">Key AI Technologies</h2>
<div class="content">
<ul class="bullet-points">
<li><span class="highlight">Machine Learning:</span> Systems that improve through experience</li>
<li><span class="highlight">Deep Learning:</span> Neural networks mimicking human brain function</li>
<li><span class="highlight">Natural Language Processing:</span> Understanding and generating human language</li>
<li><span class="highlight">Computer Vision:</span> Enabling machines to interpret visual world</li>
</ul>
</div>
</div>
<!-- Future of AI -->
<div class="slide">
<h2 class="slide-title">The Future of AI</h2>
<div class="content">
<ul class="bullet-points">
<li>🎯 Enhanced personalization in services</li>
<li>🤖 Advanced robotics and automation</li>
<li>💊 Revolutionary healthcare solutions</li>
<li>🌆 Smart cities and infrastructure</li>
<li>🌍 Environmental protection and climate solutions</li>
</ul>
</div>
</div>
<!-- Challenges -->
<div class="slide">
<h2 class="slide-title">Challenges and Considerations</h2>
<div class="content">
<ul class="bullet-points">
<li>⚖️ Ethical concerns and moral decisions</li>
<li>🔒 Privacy and data protection</li>
<li>💼 Workforce transformation and adaptation</li>
<li>⚠️ Bias and fairness in AI systems</li>
<li>🛡️ Safety and security concerns</li>
</ul>
</div>
</div>
<!-- Conclusion -->
<div class="slide">
<h2 class="slide-title">Conclusion</h2>
<div class="content">
<p>AI is transforming our world and will continue to play an increasingly important role in shaping our future.</p>
<ul class="bullet-points">
<li>🚀 Rapid advancement in technology</li>
<li>🌐 Wide-ranging impact across industries</li>
<li>🤝 Need for responsible development and governance</li>
</ul>
</div>
</div>
<script>
// Progress bar functionality
window.onscroll = function() {
let winScroll = document.body.scrollTop || document.documentElement.scrollTop;
let height = document.documentElement.scrollHeight - document.documentElement.clientHeight;
let scrolled = (winScroll / height) * 100;
document.getElementById("progressBar").style.width = scrolled + "%";
};
</script>
</body>
</html>

29
backend/agentpress/thread_manager.py
View File

@ -124,7 +124,7 @@ class ThreadManager:
# Ensure function.arguments is a string # Ensure function.arguments is a string
if 'arguments' in tool_call['function'] and not isinstance(tool_call['function']['arguments'], str): if 'arguments' in tool_call['function'] and not isinstance(tool_call['function']['arguments'], str):
# Log and fix the issue # Log and fix the issue
logger.warning(f"Found non-string arguments in tool_call, converting to string") # logger.warning(f"Found non-string arguments in tool_call, converting to string")
tool_call['function']['arguments'] = json.dumps(tool_call['function']['arguments']) tool_call['function']['arguments'] = json.dumps(tool_call['function']['arguments'])
return messages return messages
@ -146,6 +146,7 @@ class ThreadManager:
tool_choice: ToolChoice = "auto", tool_choice: ToolChoice = "auto",
native_max_auto_continues: int = 25, native_max_auto_continues: int = 25,
max_xml_tool_calls: int = 0, max_xml_tool_calls: int = 0,
include_xml_examples: bool = False,
) -> Union[Dict[str, Any], AsyncGenerator]: ) -> Union[Dict[str, Any], AsyncGenerator]:
"""Run a conversation thread with LLM integration and tool execution. """Run a conversation thread with LLM integration and tool execution.
@ -162,6 +163,7 @@ class ThreadManager:
native_max_auto_continues: Maximum number of automatic continuations when native_max_auto_continues: Maximum number of automatic continuations when
finish_reason="tool_calls" (0 disables auto-continue) finish_reason="tool_calls" (0 disables auto-continue)
max_xml_tool_calls: Maximum number of XML tool calls to allow (0 = no limit) max_xml_tool_calls: Maximum number of XML tool calls to allow (0 = no limit)
include_xml_examples: Whether to include XML tool examples in the system prompt
Returns: Returns:
An async generator yielding response chunks or error dict An async generator yielding response chunks or error dict
@ -189,6 +191,31 @@ class ThreadManager:
if max_xml_tool_calls > 0: if max_xml_tool_calls > 0:
processor_config.max_xml_tool_calls = max_xml_tool_calls processor_config.max_xml_tool_calls = max_xml_tool_calls
# Add XML examples to system prompt if requested
if include_xml_examples and processor_config.xml_tool_calling:
xml_examples = self.tool_registry.get_xml_examples()
if xml_examples:
# logger.debug(f"Adding {len(xml_examples)} XML examples to system prompt")
# Create or append to content
if isinstance(system_prompt['content'], str):
examples_content = """
In this environment you have access to a set of tools you can use to answer the user's question. The tools are specified in XML format.
{{ FORMATTING INSTRUCTIONS }}
String and scalar parameters should be specified as attributes, while content goes between tags.
Note that spaces for string values are not stripped. The output is parsed with regular expressions.
Here are the XML tools available with examples:
"""
for tag_name, example in xml_examples.items():
examples_content += f"<{tag_name}> Example: {example}\n"
system_prompt['content'] += examples_content
else:
# If content is not a string (might be a list or dict), log a warning
logger.warning("System prompt content is not a string, cannot add XML examples")
# 1. Get messages from thread for LLM call # 1. Get messages from thread for LLM call
messages = await self.get_messages(thread_id) messages = await self.get_messages(thread_id)

619
backend/poetry.lock generated
View File

File diff suppressed because one or more lines are too long

1
backend/pyproject.toml
View File

@ -46,6 +46,7 @@ python-ripgrep = "0.0.6"
daytona_sdk = "^0.12.0" daytona_sdk = "^0.12.0"
boto3 = "^1.34.0" boto3 = "^1.34.0"
openai = "^1.72.0" openai = "^1.72.0"
streamlit = "^1.44.1"
[tool.poetry.scripts] [tool.poetry.scripts]
agentpress = "agentpress.cli:main" agentpress = "agentpress.cli:main"

2
backend/utils/logger.py
View File

@ -83,7 +83,7 @@ def setup_logger(name: str = 'agentpress') -> logging.Logger:
# Console handler # Console handler
console_handler = logging.StreamHandler(sys.stdout) console_handler = logging.StreamHandler(sys.stdout)
console_handler.setLevel(logging.DEBUG) console_handler.setLevel(logging.INFO)
# Create formatters # Create formatters
file_formatter = logging.Formatter( file_formatter = logging.Formatter(