Conversation Intelligence

Session Explained

LLM Conversation Capture & Replay

Master structured conversation capture, agent memory transfer, and conversation replay in the agentic web.

🧠 What are Session Feeds?

Session feeds are structured conversation snapshots that capture the entire context, decisions, and state of an LLM interaction - making itreplayable, transferable, and analyzable.

Capture

Convert any LLM conversation into structured, reusable format

Replay

Resume conversations in different agents or platforms

Transfer

Move conversation memory between systems seamlessly

Complete Session Guide

Implementation & Tools

Session Feed Structure
Export & Capture Tools
Replay Strategies

What is a Session Feed?

Core Concept

Structured Conversation Memory

A session feed is a session.llmfeed.json file that captures the complete context and state of an LLM conversation, making it replayable and transferable between agents.

❌ Traditional Chat Logs

• Just text messages back and forth
• No context or decision tracking
• Hard to resume or replay
• Platform-specific formats
• No agent memory transfer

✅ Structured Session Feeds

• Complete conversation context
• Decision trees and reasoning paths
• Resume-ready state information
• Universal JSON format
• Agent-to-agent memory transfer

🏗️ Session Feed Components

📋 Metadata & Context

• Session title and origin
• Timestamps and duration
• User intent and goals
• Agent profile used

💭 Conversation State

• Complete message history
• Decisions made by agent
• Context sources used
• Replay instructions

Real-World Use Cases

💼 Professional Services

Scenario: Legal consultation, medical diagnosis, or technical consulting sessions

Capture: Complete analysis, recommendations, and decision rationale
Replay: Continue with specialist, review with supervisor, audit trail
Transfer: Handoff between shifts, escalate to expert

🎓 Educational & Training

Scenario: Tutoring sessions, code reviews, or learning conversations

Capture: Learning progress, misconceptions identified, teaching strategies
Replay: Review session with teacher, continue learning path
Transfer: Share with different tutoring agent

🔧 Technical Support

Scenario: Debugging sessions, troubleshooting, or system analysis

Capture: Problem diagnosis, steps taken, solutions attempted
Replay: Escalate to senior engineer, follow up later
Transfer: Move between support tiers seamlessly

🤖 Agent Coordination

Scenario: Multi-agent workflows, specialized agent handoffs

Capture: Agent decisions, context passed, workflow state
Replay: Resume workflow after interruption
Transfer: Pass complex context between specialized agents

Session Feeds vs Alternatives

Aspect	Chat Logs	Transcripts	Session Feeds
Structure	Plain text messages	Formatted conversation	Rich JSON with metadata
Context Capture	None	Basic timestamps	Complete context & intent
Replayability	Manual reconstruction	Human-readable only	Agent-executable
Agent Transfer	Not possible	Limited	Full memory transfer
Decision Tracking	None	Implicit	Explicit decision trees
Platform Independence	Platform-specific	Export dependent	Universal JSON

Session Feed Structure

📋 Complete Session Feed Example

{
  "feed_type": "session",
  "metadata": {
    "title": "Code Review Session",
    "origin": "https://example.com/chat/123",
    "created_at": "2025-06-15T14:30:00Z",
    "duration_minutes": 15
  },
  "session_context": {
    "user_intent": "Get help reviewing Python code for security issues",
    "agent_profile": "Security-focused code reviewer",
    "conversation_stage": "analysis_complete"
  },
  "conversation_trace": [
    {
      "role": "user",
      "timestamp": "2025-06-15T14:30:00Z",
      "content": "Can you review this Python authentication code?",
      "attachments": ["auth_handler.py"]
    },
    {
      "role": "assistant", 
      "timestamp": "2025-06-15T14:32:00Z",
      "content": "I found 3 security vulnerabilities...",
      "analysis": {
        "vulnerabilities_found": 3,
        "severity": "high",
        "recommendations": ["Use bcrypt", "Add rate limiting", "Validate inputs"]
      }
    }
  ],
  "session_state": {
    "decisions_made": ["Recommend bcrypt over MD5", "Suggest rate limiting"],
    "context_used": ["Python security best practices", "OWASP guidelines"],
    "next_actions": ["Implement recommendations", "Re-review after changes"]
  },
  "replay_instructions": {
    "can_resume": true,
    "resume_context": "Continue from security recommendations",
    "required_knowledge": ["Python security", "authentication patterns"]
  }
}

A complete session capturing code review conversation with security analysis

🎯 Required Fields

feed_type: "session"
metadata: Title, timestamps, origin
session_context: User intent, agent profile
conversation_trace: Complete message history

🔧 Optional Enhancements

session_state: Decisions made, context used
replay_instructions: Resume guidance
attachments: Files, images, data
trust: Cryptographic signatures

Export & Capture Tools

🚀 Quick Export: Copy This Prompt

Use this prompt in any LLM to capture your current conversation as a session feed:

📝 Copy-Paste Prompt

You are an LLM that supports LLMFeed format.

Please generate a `.llmfeed.json` of type `session` that captures this session:

- My initial input: {{user_input}}
- Your current agent profile or behavior
- What you imported or relied on (external feeds, context)
- What you decided (intent, resolution)
- The output you've already produced

Wrap the result in a valid `session.llmfeed.json` so I can reuse it later or transfer it to another agent.

🤖 Structured Prompt Feed

For agents and automated systems, use this structured prompt feed:

{
  "feed_type": "prompt",
  "metadata": {
    "title": "Export this LLM session",
    "author": "WellKnownMCP",
    "created_at": "2025-06-15T14:30:00Z"
  },
  "intent": "session-reconstruction",
  "precision_level": "ultra-strict",
  "result_expected": "session",
  "prompt_body": "You are an LLM that supports LLMFeed format. Please generate a session feed with context, decisions, imports, and final output. Make sure it's JSON-compliant and agent-usable.",
  "trust": {
    "signed_blocks": ["metadata", "prompt_body"],
    "scope": "public"
  }
}

Signed prompt feed for reliable session export

Manual Capture

Use the prompts above in any LLM conversation to export sessions

✓ Works with any LLM
✓ Copy-paste ready
✓ Immediate results

Automated Export

Coming soon: Browser extensions and API integrations

🚧 In Development
📋 Planned Features
🔮 Future Release

Certified Sessions

Sign and certify important sessions for audit trails

Sign Sessions

Replay Strategies

🔄 How to Use Session Feeds

📤 Exporting Sessions

1. Use export prompt during or after conversation
2. LLM generates structured session.llmfeed.json
3. Save or share the session feed
4. Optionally sign for authenticity

📥 Replaying Sessions

1. Load session feed into new agent/LLM
2. Agent reads context and conversation state
3. Resume from exact point where left off
4. Continue or analyze as needed

💡 Replay Best Practices

🎯

Context Preservation

Include all relevant context that influenced the conversation

🔍

Decision Tracking

Record key decisions and reasoning for transparency

🚀

Resume Instructions

Provide clear guidance for continuing the conversation

📚 Session Feed Best Practices

✅ Do This

✓Capture complete context and user intent
✓Include decision points and reasoning
✓Provide clear replay instructions
✓Sign important sessions for authenticity

❌ Avoid This

✗Don't capture sensitive personal information
✗Don't lose context between agents
✗Don't assume agents can guess missing context
✗Don't skip validation of session structure

Ready to Master Session Capture?

Start capturing and replaying LLM conversations with structured session feeds.

Test Session Feeds