Skip to main content
Session summary provides a token-efficient way to include task context in your agent’s prompts. Instead of manually iterating through tasks and formatting them, use get_session_summary to get a pre-formatted string ready for prompt injection.

​
Why Use Session Summary?

When building agents that need awareness of their previous work, you often want to include task history in the system prompt or context. The session summary method:
  • Saves tokens: Compact format optimized for LLM consumption
  • Let your agent run longer: Consider this summary is a automatic TODO list, your agent can use it to track its work and progress.

​
Usage

from acontext import AcontextClient

client = AcontextClient()

# Get session summary
summary = client.sessions.get_session_summary(session_id)

# Use in your prompt
system_prompt = f"""You are a helpful assistant.

Previous tasks in this session:
{summary}

Continue helping the user with their request.
"""
Task extraction runs asynchronously, so the session summary may lag behind by 1-6 messages. If you’re using session summary as a compression method, keep at least the 4-6 most recent messages in full to maintain consistency between the summary and the conversation context.

​
Output Format

The summary returns an XML-formatted string optimized for LLM parsing: 
<task id="1" description="Search for the latest news about iPhone 15 pro max">
<progress>
1. Searched and found key specifications
2. Compiled feature comparison
</progress>
<user_preference>
1. Focus on camera capabilities
2. Make it mobile-responsive
</user_preference>
</task>
<task id="2" description="Initialize Next.js project">
<progress>
1. Created project with latest template
</progress>
</task>
Each task element includes:
  • id attribute - Task order number
  • description attribute - Task description
  • <progress> element with numbered steps (if any)
  • <user_preference> element with numbered items (if any)

​
How It Works

Session summary leverages Acontext’s automatic task extraction which runs in the background as your agent converses. The extracted tasks include:
  • Task description: What the agent planned to do
  • Progress updates: Steps the agent completed
  • User preferences: Requirements mentioned by the user
The get_session_summary method fetches these tasks and formats them into a compact string.

​
Options

You can limit the number of tasks returned: 
# Get only the last 5 tasks
summary = client.sessions.get_session_summary(session_id, limit=5)
If no tasks have been extracted yet, the method returns an empty string. Task extraction happens asynchronously - see Session Buffer for details on timing.

​
Next Steps

Task Extraction

Learn how Acontext automatically extracts tasks from conversations.

Context Editing

Apply editing strategies to manage context window size.