Skip to main content

Feed Type: session.llmfeed.json

Purpose​

A session.llmfeed.json captures the context and progress of a conversation with an LLM so you can continue it seamlessly with another LLM or resume it later.

Think of it as "copy-paste on steroids" - instead of losing context when switching between Claude, ChatGPT, or other LLMs, you get perfect continuity.

The Original Problem​

What usually happens:

You: [Long conversation with Claude about React optimization]
Claude: [Helpful analysis and recommendations]
You: [Switch to ChatGPT]
ChatGPT: "I don't have context of your previous conversation..."
You: [Have to re-explain everything] 😞

What session feeds enable:

You: "Claude, export this conversation for me to continue elsewhere"
Claude: [Generates session.llmfeed.json with full context]
You: [Paste into ChatGPT] 
ChatGPT: "I see you were discussing React optimization with Claude. 
         You're implementing custom hooks for data fetching..." 😊

Minimum Structure (Hello World)​

{
  "feed_type": "session",
  "metadata": {
    "title": "React Optimization Discussion",
    "origin": "claude.ai",
    "generated_at": "2025-06-10T15:30:00Z"
  },
  "session_summary": {
    "topic": "React performance optimization",
    "current_focus": "custom hooks for data fetching",
    "key_points": ["useState vs useRef", "memoization patterns"]
  }
}

Result: Any LLM can understand where you left off!

Common Use Cases​

Switching Between LLMs​

{
  "feed_type": "session",
  "metadata": {
    "title": "Writing Strategy Session",
    "origin": "claude.ai",
    "generated_at": "2025-06-10T14:00:00Z"
  },
  "session_summary": {
    "topic": "Blog post about sustainable technology",
    "current_focus": "structuring the article outline",
    "key_decisions": ["target audience: tech professionals", "tone: informative but engaging"],
    "next_steps": "need help with compelling introduction"
  },
  "conversation_context": {
    "user_background": "startup founder",
    "writing_style": "prefers short paragraphs",
    "target_length": "1500 words"
  }
}

Pause and Resume​

{
  "feed_type": "session", 
  "metadata": {
    "title": "Python Learning Session",
    "origin": "chatgpt.com",
    "generated_at": "2025-06-10T16:45:00Z"
  },
  "session_summary": {
    "topic": "Learning Python decorators",
    "current_focus": "understanding @property decorator",
    "completed_topics": ["basic decorators", "function wrappers"],
    "next_steps": "practice with class decorators"
  },
  "learning_context": {
    "skill_level": "intermediate",
    "learning_goal": "build a web scraper",
    "time_constraint": "2 weeks to complete project"
  }
}

Collaborative Sessions​

{
  "feed_type": "session",
  "metadata": {
    "title": "Team Planning Session",
    "origin": "gemini.google.com", 
    "generated_at": "2025-06-10T11:20:00Z"
  },
  "session_summary": {
    "topic": "Q3 product roadmap planning",
    "participants": ["product_manager", "tech_lead", "designer"],
    "current_focus": "prioritizing feature requests",
    "decisions_made": ["mobile app is priority", "API v2 delayed to Q4"]
  },
  "project_context": {
    "team_size": "8 people",
    "deadline": "end of Q3",
    "budget_constraints": "limited engineering resources"
  }
}

Trust & Verification​

Basic Signed Session​

{
  "feed_type": "session",
  "metadata": {
    "title": "Medical Research Discussion",
    "origin": "claude.ai",
    "generated_at": "2025-06-10T13:15:00Z"
  },
  "session_summary": {
    "topic": "diabetes treatment research",
    "current_focus": "analyzing recent clinical trials"
  },
  "trust": {
    "signed_blocks": ["metadata", "session_summary", "trust"],
    "scope": "conversation_integrity"
  },
  "signature": {
    "value": "signature_hash_here",
    "created_at": "2025-06-10T13:15:00Z"
  }
}

Why sign sessions?

  • ✅ Prove authenticity - verify the session really came from that LLM
  • ✅ Prevent tampering - ensure context wasn't modified
  • ✅ Build trust - especially important for sensitive topics
  • ✅ Enable audit trails - track conversation history

Core Fields​

FieldRequiredDescription
session_summary✅What was discussed and current status
conversation_context⚠️User background and preferences
key_decisions⚠️Important choices or conclusions made
next_steps⚠️What to focus on next
completed_topics⚠️What's already been covered

How LLMs Should Handle Sessions​

When Receiving a Session Feed​

  1. Read the context - understand what was discussed
  2. Acknowledge continuity - mention the previous conversation
  3. Pick up where left off - focus on current_focus and next_steps
  4. Respect the style - adapt to user preferences in conversation_context

Example Response​

"I can see you were discussing React optimization with Claude, and you're 
currently working on custom hooks for data fetching. Based on your preference 
for functional components and your e-commerce dashboard project, let's continue 
by looking at some practical patterns for data fetching hooks..."

When Asked to Export​

  1. Summarize the conversation - key topics and decisions
  2. Capture current focus - where the discussion stands
  3. Include user context - background and preferences revealed
  4. Suggest next steps - logical continuation points

📚 Advanced Features​

Decision Tracking

Conversation with Decision Points​

{
  "session_summary": {
    "topic": "choosing a web framework",
    "current_focus": "comparing React vs Vue",
    "key_decisions": [
      {
        "decision": "ruled out Angular",
        "reasoning": "too complex for team size",
        "timestamp": "2025-06-10T14:20:00Z"
      },
      {
        "decision": "prioritizing ease of learning", 
        "reasoning": "team has junior developers",
        "timestamp": "2025-06-10T14:25:00Z"
      }
    ]
  }
}
Multi-Session Workflows
{
  "related_sessions": [
    {
      "title": "Initial Project Planning",
      "url": "/sessions/project-planning-20250605.llmfeed.json",
      "relationship": "precedes"
    },
    {
      "title": "Technical Deep Dive", 
      "url": "/sessions/tech-deep-dive-20250608.llmfeed.json",
      "relationship": "parallel_track"
    }
  ]
}
Learning and Adaptation

Educational Sessions​

{
  "learning_progress": {
    "concepts_mastered": ["Python basics", "loops", "functions"],
    "current_difficulty": "object-oriented programming",
    "learning_style": "prefers examples over theory",
    "misconceptions_corrected": [
      "thought classes were like functions",
      "confused about self parameter"
    ]
  }
}

Professional Development​

{
  "career_context": {
    "current_role": "junior frontend developer",
    "learning_goals": ["become full-stack", "understand backend"],
    "timeline": "6 months",
    "preferred_technologies": ["JavaScript", "Python", "PostgreSQL"]
  }
}
Collaborative and Team Sessions

Team Planning​

{
  "collaboration_context": {
    "session_type": "team_planning",
    "facilitator": "product_manager",
    "stakeholders": ["engineering", "design", "marketing"],
    "consensus_points": ["mobile-first approach", "Q3 launch target"],
    "open_questions": ["budget allocation", "resource allocation"]
  }
}

Multi-Agent Workflows​

{
  "agent_collaboration": {
    "primary_agent": "claude.ai",
    "consulting_agents": ["code_review_bot", "design_feedback_ai"],
    "handoff_points": [
      {
        "to_agent": "code_review_bot",
        "for_task": "review generated React components",
        "expected_return": "code quality feedback"
      }
    ]
  }
}

Best Practices​

For Users​

  1. Export at natural break points - end of topics, before switching focus
  2. Include enough context - background info helps the new LLM understand
  3. Be specific about next steps - what you want to accomplish next
  4. Use descriptive titles - makes sessions easier to find later

For LLM Implementations​

  1. Always acknowledge the handoff - mention the previous conversation
  2. Summarize what you understand - confirm context was preserved
  3. Pick up naturally - don't restart, continue the flow
  4. Respect user preferences - adapt to their communication style

For Developers​

  1. Keep sessions focused - one session per major topic/goal
  2. Sign important sessions - especially for sensitive or critical content
  3. Link related sessions - create workflows across multiple conversations
  4. Include user context - background helps with personalization

Real-World Examples​

Code Review Session​

{
  "feed_type": "session",
  "metadata": {
    "title": "React Component Code Review",
    "origin": "claude.ai"
  },
  "session_summary": {
    "topic": "reviewing custom React components",
    "current_focus": "optimizing render performance",
    "code_reviewed": ["UserProfile.jsx", "ProductCard.jsx"],
    "issues_found": ["missing memo optimization", "prop drilling"],
    "next_steps": "implement context API for state management"
  },
  "development_context": {
    "project_type": "e-commerce dashboard",
    "tech_stack": ["React 18", "TypeScript", "Vite"],
    "performance_requirements": "sub-200ms render times"
  }
}

Writing Workshop Session​

{
  "feed_type": "session",
  "metadata": {
    "title": "Technical Blog Writing Session",
    "origin": "chatgpt.com"
  },
  "session_summary": {
    "topic": "writing about machine learning concepts",
    "current_focus": "making complex topics accessible",
    "completed_sections": ["introduction", "basic concepts"],
    "writing_challenges": ["explaining neural networks simply"],
    "next_steps": "add practical examples and code snippets"
  },
  "writing_context": {
    "target_audience": "junior developers",
    "publication": "company engineering blog",
    "tone": "educational but conversational",
    "length_goal": "2000 words"
  }
}

What This Enables​

For Users​

  • ✅ Seamless LLM switching - no context loss
  • ✅ Better conversation continuity - pick up exactly where you left off
  • ✅ Portable knowledge sessions - take your conversations anywhere
  • ✅ Long-term project tracking - maintain context across weeks/months

For LLMs​

  • ✅ Better user experience - understanding context immediately
  • ✅ More effective assistance - building on previous work
  • ✅ Reduced repetition - no need to re-establish background
  • ✅ Collaborative workflows - working together across platforms

For Teams​

  • ✅ Shared context - team members can continue each other's conversations
  • ✅ Knowledge preservation - important discussions don't get lost
  • ✅ Decision tracking - clear record of choices and reasoning
  • ✅ Workflow continuity - projects span multiple sessions and people

MIME Type​

Content-Type: application/llmfeed+json

Or specifically:

Content-Type: application/session+llmfeed
  • export.llmfeed.json: Export specific content from a session
  • prompt.llmfeed.json: Reusable prompts that work across sessions
  • mcp.llmfeed.json: Service context that sessions can reference
  • credential.llmfeed.json: Authentication for private session content

References​