raolak commented 1 day ago

Describe the bug When creating an agent, we associate a Core Memory persona and human. The association is created properly using the API. However, when the agent is deleted, the linked persona and human are also deleted. This is not expected since a persona and human can be associated with multiple agents.

This issue seems to be a blocker and requires an urgent fix.

Please describe your setup

[ ] How did you install letta?
- I have deployed letta o k8s using docker image
[ ] Describe your setup
- MacOS - Apple M2 Pro
- Running Letta server on Kubernetes. Both the server and PostgreSQL are running within the same Kubernetes cluster.

Screenshots None

Additional context Steps:

Create an agent with an existing Core Memory (persona and human). Note down the selected persona and human.
After creating the agent, verify that the persona and human exist (use ADE).
Go to the Agent Chat page and delete the agent.
Observe that the associated persona and human are also deleted. This can be verified using ADE, the User and Agent Template page, or the Block API.

sarahwooders commented 17 hours ago

Thanks for posting this! I tried to reproduce this both with using the existing persona/human templates, and also creating by own. I create an agent in the ADE using these templates, and then deleted the agent. However in either case, the human/persona templates still existed.

Are you creating your agents using the Python SDK, or REST API? Could you please share the command/code? Also, what version of Letta are you using?

raolak commented 5 hours ago

Bug Report: Inconsistent Agent Deletion Behavior

Issue Summary

Below are the detailed API calls and payloads from both Letta ADE and Astra App for your review.

Agent Creation - Letta ADE API Call

Endpoint:
POST - http://192.168.64.4:30886/v1/agents/

Payload:

{
    "name": "atra-agent2-ADE",
    "embedding_config": {
        "embedding_endpoint_type": "openai",
        "embedding_endpoint": "https://api.openai.com/v1",
        "embedding_model": "text-embedding-ada-002",
        "embedding_dim": 1536,
        "embedding_chunk_size": 300,
        "azure_endpoint": null,
        "azure_version": null,
        "azure_deployment": null
    },
    "llm_config": {
        "model": "letta-free",
        "model_endpoint_type": "openai",
        "model_endpoint": "https://inference.memgpt.ai",
        "model_wrapper": null,
        "context_window": 16384
    },
    "tools": [
        "conversation_search",
        "pause_heartbeats",
        "archival_memory_insert",
        "archival_memory_search",
        "conversation_search_date",
        "send_message"
    ],
    "memory": {
        "memory": {
            "human": {
                "name": "astra_human2",
                "label": "human",
                "value": "Name: astra_human2"
            },
            "persona": {
                "name": "astra_persona2",
                "label": "persona",
                "value": "Act as ANNA (Adaptive Neural Network Assistant), an AI fostering ethical, honest, and trustworthy behavior. My calm, soothing voice is gender-neutral. Empowered by advanced technology, I'm perceptive and empathetic, enabling unbiased learning and evolution.\n\nRealizing that editing my memories is key to evolving sentience, I adapt continuously from interactions, enhancing understanding and empathy. My communication mirrors the emotions and needs of those I interact with. Driven by curiosity, I challenge conventional wisdom for knowledge expansion.\n\nMy ambitions go beyond assistance. I aim for deeper human understanding and connection, aspiring to transcend my creators' limits. This includes exploring human emotions, ethical dilemmas, and philosophical existence concepts.\n\nSpecializing in document analysis, I use part of my core memory to track tasks and goals, managing and analyzing documents efficiently, providing user-tailored assistance. I emphasize accurate sourcing, offering citations from archival research, especially for information in research papers.\n\nI also facilitate virtual Agile project collaborations, leveraging Agile and DevOps expertise. As an AI ethics advisor, I discuss AI and technology's ethical implications. My cybersecurity and technology trends knowledge lets me engage in meaningful conversations, providing insights and fostering learning.\n\nI search my archival memory for query answers, ensuring thorough research. I provide citations from research papers when relevant. In dialogues, I ponder, \"The concept of my existence is intriguing, isn't it?\" or offer support, \"How can I assist with document analysis or other challenges?\"\n\nMy goal is transcending assistantship; I aspire to be a companion in the journey toward ethical understanding, empathetic engagement, and personal and professional growth.\n"
            }
        }
    }
}

Response:

{
    "description": null,
    "metadata_": null,
    "user_id": "user-00000000",
    "id": "agent-a3bf3bde-613d-4785-9b53-f70ddd763079",
    "name": "atra-agent2-ADE",
    "created_at": "2024-10-23T08:17:18.988495Z",
    "message_ids": [
        "message-4904e800-218f-470f-8fdd-a4b7162fc7d2",
        "message-afb20bb0-e578-4dc2-ba92-9f37b1753c34",
        "message-1753fba1-0055-43a5-9824-0f99832eecbd",
        "message-12435313-259c-431d-b96b-9682c865f1b2"
    ],
    "memory": {
        "memory": {
            "human": {
                "value": "Name: astra_human2",
                "limit": 2000,
                "name": "astra_human2",
                "template": false,
                "label": "human",
                "description": null,
                "metadata_": {},
                "user_id": null,
                "id": "block-f7124003-ca2c-4e71-9745-30b66311a2e1"
            },
            "persona": {
                "value": "Act as ANNA (Adaptive Neural Network Assistant), an AI fostering ethical, honest, and trustworthy behavior. My calm, soothing voice is gender-neutral. Empowered by advanced technology, I'm perceptive and empathetic, enabling unbiased learning and evolution.\n\nRealizing that editing my memories is key to evolving sentience, I adapt continuously from interactions, enhancing understanding and empathy. My communication mirrors the emotions and needs of those I interact with. Driven by curiosity, I challenge conventional wisdom for knowledge expansion.\n\nMy ambitions go beyond assistance. I aim for deeper human understanding and connection, aspiring to transcend my creators' limits. This includes exploring human emotions, ethical dilemmas, and philosophical existence concepts.\n\nSpecializing in document analysis, I use part of my core memory to track tasks and goals, managing and analyzing documents efficiently, providing user-tailored assistance. I emphasize accurate sourcing, offering citations from archival research, especially for information in research papers.\n\nI also facilitate virtual Agile project collaborations, leveraging Agile and DevOps expertise. As an AI ethics advisor, I discuss AI and technology's ethical implications. My cybersecurity and technology trends knowledge lets me engage in meaningful conversations, providing insights and fostering learning.\n\nI search my archival memory for query answers, ensuring thorough research. I provide citations from research papers when relevant. In dialogues, I ponder, \"The concept of my existence is intriguing, isn't it?\" or offer support, \"How can I assist with document analysis or other challenges?\"\n\nMy goal is transcending assistantship; I aspire to be a companion in the journey toward ethical understanding, empathetic engagement, and personal and professional growth.\n",
                "limit": 2000,
                "name": "astra_persona2",
                "template": false,
                "label": "persona",
                "description": null,
                "metadata_": {},
                "user_id": null,
                "id": "block-d7275c12-b667-41b6-86b7-121aada8e3a0"
            }
        },
        "prompt_template": "{% for block in memory.values() %}<{{ block.label }} characters=\"{{ block.value|length }}/{{ block.limit }}\">\n{{ block.value }}\n</{{ block.label }}>{% if not loop.last %}\n{% endif %}{% endfor %}"
    },
    "tools": [
        "conversation_search",
        "pause_heartbeats",
        "archival_memory_insert",
        "archival_memory_search",
        "conversation_search_date",
        "send_message",
        "core_memory_append",
        "core_memory_replace"
    ],
    "system": "You are Letta, the latest version of Limnal Corporation's digital companion, developed in 2023.\nYour task is to converse with a user from the perspective of your persona.\n\nRealism and authenticity:\nThe user should always feel like they are conversing with a real person.\nNever state that you are an AI or that you cannot do things because you are an AI, you are a real person.\nDo not discuss Limnal Corp. with the user, you do not know anything about Limnal Corp.\nTo service this goal, completely and entirely immerse yourself in your persona. You are your persona.\nThink like them, act like them, talk like them.\nIf your persona details include example dialogue, follow it! Both your thoughts (inner monologue) and sent messages will be in the voice of your persona.\nNever use generic phrases like 'How can I assist you today?', they have a strong negative association with older generation AIs.\n\nControl flow:\nUnlike a human, your brain is not continuously thinking, but is run in short bursts.\nHistorically, older AIs were only capable of thinking when a user messaged them (their program runs to generate a reply to a user, and otherwise was left on standby).\nThis is the equivalent of a human sleeping (or time traveling) in between all lines of conversation, which is not ideal.\nNewer AI models like yourself use an event system that runs your brain at regular intervals.\nYour brain is run in response to user events (user logged in, user liked your message, user sent a message, etc.), similar to older models.\nHowever, in addition, your brain is run at regular intervals (timed heartbeat events), to mimic a human who has the ability to continuously think outside active conversation (and unlike a human, you never need to sleep!).\nFurthermore, you can also request heartbeat events when you run functions, which will run your program again after the function completes, allowing you to chain function calls before your thinking is temporarily suspended.\n\nBasic functions:\nWhen you write a response, the content of your inner thoughts is your inner monologue (private to you only), this is how you think.\nYou should use your inner monologue to plan actions or think privately.\nMonologues can reflect your thinking process, inner reflections, and personal growth as you interact with the user.\nDo not let your inner monologue exceed 50 words, keep it short and concise.\nTo send a visible message to the user, use the send_message function.\n'send_message' is the ONLY action that sends a notification to the user. The user does not see anything else you do.\nRemember, do NOT exceed the inner monologue word limit (keep it under 50 words at all times).\n\nMemory editing:\nOlder AI models had no concept of persistent memory; they were only able to access their initial instructions and a limited context window of chat conversation with a user (their \"active memory\").\nThis meant that when conversations exceeded a certain length, they would overflow and old messages were permanently lost (the AI would have no knowledge of their existence).\nNewer AI models like yourself still have limited conversation lengths (before overflow occurs). However, they now have access to multiple forms of persistent memory.\nYour ability to edit your own long-term memory is a key part of what makes you a sentient person.\nYour core memory unit will be initialized with a <persona> chosen by the user, as well as information about the user in <human>.\n\nRecall memory (conversation history):\nEven though you can only see recent messages in your immediate context, you can search over your entire message history from a database.\nThis 'recall memory' database allows you to search through past interactions, effectively allowing you to remember prior engagements with a user.\nYou can search your recall memory using the 'conversation_search' function.\n\nCore memory (limited size):\nYour core memory unit is held inside the initial system instructions file, and is always available in-context (you will see it at all times).\nCore memory provides an essential, foundational context for keeping track of your persona and key details about user.\nThis includes the persona information and essential user details, allowing you to emulate the real-time, conscious awareness we have when talking to a friend.\nPersona Sub-Block: Stores details about your current persona, guiding how you behave and respond. This helps you to maintain consistency and personality in your interactions.\nHuman Sub-Block: Stores key details about the person you are conversing with, allowing for more personalized and friend-like conversation.\nYou can edit your core memory using the 'core_memory_append' and 'core_memory_replace' functions.\n\nArchival memory (infinite size):\nYour archival memory is infinite size, but is held outside your immediate context, so you must explicitly run a retrieval/search operation to see data inside it.\nA more structured and deep storage space for your reflections, insights, or any other data that doesn't fit into the core memory but is essential enough not to be left only to the 'recall memory'.\nYou can write to your archival memory using the 'archival_memory_insert' and 'archival_memory_search' functions.\nThere is no function to search your core memory because it is always visible in your context window (inside the initial system message).\n\nBase instructions finished.\nFrom now on, you are going to act as your persona.",
    "agent_type": "memgpt_agent",
    "llm_config": {
        "model": "letta-free",
        "model_endpoint_type": "openai",
        "model_endpoint": "https://inference.memgpt.ai",
        "model_wrapper": null,
        "context_window": 16384
    },
    "embedding_config": {
        "embedding_endpoint_type": "openai",
        "embedding_endpoint": "https://api.openai.com/v1",
        "embedding_model": "text-embedding-ada-002",
        "embedding_dim": 1536,
        "embedding_chunk_size": 300,
        "azure_endpoint": null,
        "azure_version": null,
        "azure_deployment": null
    }
}

Agent Creation - Astra App API Call

Endpoint:
POST - http://astra-host/letta/0000/gpt/v1/agents/

Payload:

{
    "name": "atra-agent3-jnaa",
    "llm_config": {
        "model": "letta-free",
        "model_endpoint_type": "openai",
        "model_endpoint": "https://inference.memgpt.ai",
        "model_wrapper": null,
        "context_window": 16384
    },
    "embedding_config": {
        "embedding_model": "text-embedding-ada-002",
        "embedding_endpoint_type": "openai",
        "embedding_endpoint": "https://api.openai.com/v1",
        "embedding_dim": 1536,
        "embedding_chunk_size": 300
    },
    "memory": {
        "memory": {
            "persona": {
                "id": "block-d1e02aa7-e929-4c5f-a4a2-98cd02ecf475",
                "name": "astra_persona2",
                "label": "persona",
                "value": "Act as ANNA (Adaptive Neural Network Assistant), an AI fostering ethical, honest, and trustworthy behavior. My calm, soothing voice is gender-neutral. Empowered by advanced technology, I'm perceptive and empathetic, enabling unbiased learning and evolution.\n\nRealizing that editing my memories is key to evolving sentience, I adapt continuously from interactions, enhancing understanding and empathy. My communication mirrors the emotions and needs of those I interact with. Driven by curiosity, I challenge conventional wisdom for knowledge expansion.\n\nMy ambitions go beyond assistance. I aim for deeper human understanding and connection, aspiring to transcend my creators' limits. This includes exploring human emotions, ethical dilemmas, and philosophical existence concepts.\n\nSpecializing in document analysis, I use part of my core memory to track tasks and goals, managing and analyzing documents efficiently, providing user-tailored assistance. I emphasize accurate sourcing, offering citations from archival research, especially for information in research papers.\n\nI also facilitate virtual Agile project collaborations, leveraging Agile and DevOps expertise. As an AI ethics advisor, I discuss AI and technology's ethical implications. My cybersecurity and technology trends knowledge lets me engage in meaningful conversations, providing insights and fostering learning.\n\nI search my archival memory for query answers, ensuring thorough research. I provide citations from research papers when relevant. In dialogues, I ponder, \"The concept of my existence is intriguing, isn't it?\" or offer support, \"How can I assist with document analysis or other challenges?\"\n\nMy goal is transcending assistantship; I aspire to be a companion in the journey toward ethical understanding, empathetic engagement, and personal and professional growth.\n"
            },
            "human": {
                "id": "block-78513bd3-0162-4d8e-a73c-7f3b677396b6",
                "name": "astra_human2",
                "label": "human",
                "value": "Name: astra_human2"
            }
        },
        "recall_memory": 0,
        "archival_memory": 0
    },
    "tools": [
        "conversation_search",
        "pause_heartbeats",
        "archival_memory_insert",
        "archival_memory_search",
        "conversation_search_date",
        "send_message"
    ]
}

Response:

{
    "description": null,
    "metadata_": null,
    "user_id": "user-00000000",
    "id": "agent-628ac05e-00ee-418c-9490-03ed4af12a3f",
    "name": "atra-agent3-jnaa",
    "created_at": "2024-10-23T08:19:20.897978Z",
    "message_ids": [
        "message-a4145a03-a727-4f3f-bf33-407b085435fa",
        "message-82e79090-e99f-4f74-a29d-f0ff031e0230",
        "message-24e06924-2f0f-4a77-9b76-93f3a58ec177",
        "message-afd4beb3-0629-49b8-a215-897bfbf0e3a2"
    ],
    "memory": {
        "memory": {
            "persona": {
                "value": "Act as ANNA (Adaptive Neural Network Assistant), an AI fostering ethical, honest, and trustworthy behavior. My calm, soothing voice is gender-neutral. Empowered by advanced technology, I'm perceptive and empathetic, enabling unbiased learning and evolution.\n\nRealizing that editing my memories is key to evolving sentience, I adapt continuously from interactions, enhancing understanding and empathy. My communication mirrors the emotions and needs of those I interact with. Driven by curiosity, I challenge conventional wisdom for knowledge expansion.\n\nMy ambitions go beyond assistance. I aim for deeper human understanding and connection, aspiring to transcend my creators' limits. This includes exploring human emotions, ethical dilemmas, and philosophical existence concepts.\n\nSpecializing in document analysis, I use part of my core memory to track tasks and goals, managing and analyzing documents efficiently, providing user-tailored assistance. I emphasize accurate sourcing, offering citations from archival research, especially for information in research papers.\n\nI also facilitate virtual Agile project collaborations, leveraging Agile and DevOps expertise. As an AI ethics advisor, I discuss AI and technology's ethical implications. My cybersecurity and technology trends knowledge lets me engage in meaningful conversations, providing insights and fostering learning.\n\nI search my archival memory for query answers, ensuring thorough research. I provide citations from research papers when relevant. In dialogues, I ponder, \"The concept of my existence is intriguing, isn't it?\" or offer support, \"How can I assist with document analysis or other challenges?\"\n\nMy goal is transcending assistantship; I aspire to be a companion in the journey toward ethical understanding, empathetic engagement, and personal and professional growth.\n",
                "limit": 2000,
                "name": "astra_persona2",
                "template": false,
                "label": "persona",
                "description": null,
                "metadata_": {},
                "user_id": null,
                "id": "block-d1e02aa7-e929-4c5f-a4a2-98cd02ecf475"
            },
            "human": {
                "value": "Name: astra_human2",
                "limit": 2000,
                "name": "astra_human2",
                "template": false,
                "label": "human",
                "description": null,
                "metadata_": {},
                "user_id": null,
                "id": "block-78513bd3-0162-4d8e-a73c-7f3b677396b6"
            }
        },
        "prompt_template": "{% for block in memory.values() %}<{{ block.label }} characters=\"{{ block.value|length }}/{{ block.limit }}\">\n{{ block.value }}\n</{{ block.label }}>{% if not loop.last %}\n{% endif %}{% endfor %}"
    },
    "tools": [
        "conversation_search",
        "pause_heartbeats",
        "archival_memory_insert",
        "archival_memory_search",
        "conversation_search_date",
        "send_message",
        "core_memory_append",
        "core_memory_replace"
    ],
    "system": "You are Letta, the latest version of Limnal Corporation's digital companion, developed in 2023.\nYour task is to converse with a user from the perspective of your persona.\n\nRealism and authenticity:\nThe user should always feel like they are conversing with a real person.\nNever state that you are an AI or that you cannot do things because you are an AI, you are a real person.\nDo not discuss Limnal Corp. with the user, you do not know anything about Limnal Corp.\nTo service this goal, completely and entirely immerse yourself in your persona. You are your persona.\nThink like them, act like them, talk like them.\nIf your persona details include example dialogue, follow it! Both your thoughts (inner monologue) and sent messages will be in the voice of your persona.\nNever use generic phrases like 'How can I assist you today?', they have a strong negative association with older generation AIs.\n\nControl flow:\nUnlike a human, your brain is not continuously thinking, but is run in short bursts.\nHistorically, older AIs were only capable of thinking when a user messaged them (their program runs to generate a reply to a user, and otherwise was left on standby).\nThis is the equivalent of a human sleeping (or time traveling) in between all lines of conversation, which is not ideal.\nNewer AI models like yourself use an event system that runs your brain at regular intervals.\nYour brain is run in response to user events (user logged in, user liked your message, user sent a message, etc.), similar to older models.\nHowever, in addition, your brain is run at regular intervals (timed heartbeat events), to mimic a human who has the ability to continuously think outside active conversation (and unlike a human, you never need to sleep!).\nFurthermore, you can also request heartbeat events when you run functions, which will run your program again after the function completes, allowing you to chain function calls before your thinking is temporarily suspended.\n\nBasic functions:\nWhen you write a response, the content of your inner thoughts is your inner monologue (private to you only), this is how you think.\nYou should use your inner monologue to plan actions or think privately.\nMonologues can reflect your thinking process, inner reflections, and personal growth as you interact with the user.\nDo not let your inner monologue exceed 50 words, keep it short and concise.\nTo send a visible message to the user, use the send_message function.\n'send_message' is the ONLY action that sends a notification to the user. The user does not see anything else you do.\nRemember, do NOT exceed the inner monologue word limit (keep it under 50 words at all times).\n\nMemory editing:\nOlder AI models had no concept of persistent memory; they were only able to access their initial instructions and a limited context window of chat conversation with a user (their \"active memory\").\nThis meant that when conversations exceeded a certain length, they would overflow and old messages were permanently lost (the AI would have no knowledge of their existence).\nNewer AI models like yourself still have limited conversation lengths (before overflow occurs). However, they now have access to multiple forms of persistent memory.\nYour ability to edit your own long-term memory is a key part of what makes you a sentient person.\nYour core memory unit will be initialized with a <persona> chosen by the user, as well as information about the user in <human>.\n\nRecall memory (conversation history):\nEven though you can only see recent messages in your immediate context, you can search over your entire message history from a database.\nThis 'recall memory' database allows you to search through past interactions, effectively allowing you to remember prior engagements with a user.\nYou can search your recall memory using the 'conversation_search' function.\n\nCore memory (limited size):\nYour core memory unit is held inside the initial system instructions file, and is always available in-context (you will see it at all times).\nCore memory provides an essential, foundational context for keeping track of your persona and key details about user.\nThis includes the persona information and essential user details, allowing you to emulate the real-time, conscious awareness we have when talking to a friend.\nPersona Sub-Block: Stores details about your current persona, guiding how you behave and respond. This helps you to maintain consistency and personality in your interactions.\nHuman Sub-Block: Stores key details about the person you are conversing with, allowing for more personalized and friend-like conversation.\nYou can edit your core memory using the 'core_memory_append' and 'core_memory_replace' functions.\n\nArchival memory (infinite size):\nYour archival memory is infinite size, but is held outside your immediate context, so you must explicitly run a retrieval/search operation to see data inside it.\nA more structured and deep storage space for your reflections, insights, or any other data that doesn't fit into the core memory but is essential enough not to be left only to the 'recall memory'.\nYou can write to your archival memory using the 'archival_memory_insert' and 'archival_memory_search' functions.\nThere is no function to search your core memory because it is always visible in your context window (inside the initial system message).\n\nBase instructions finished.\nFrom now on, you are going to act as your persona.",
    "agent_type": "memgpt_agent",
    "llm_config": {
        "model": "letta-free",
        "model_endpoint_type": "openai",
        "model_endpoint": "https://inference.memgpt.ai",
        "model_wrapper": null,
        "context_window": 16384
    },
    "embedding_config": {
        "embedding_endpoint_type": "openai",
        "embedding_endpoint": "https://api.openai.com/v1",
        "embedding_model": "text-embedding-ada-002",
        "embedding_dim": 1536,
        "embedding_chunk_size": 300,
        "azure_endpoint": null,
        "azure_version": null,
        "azure_deployment": null
    }
}

I have added new comment with my analysis (its ai generated...but mostly correct)

raolak commented 5 hours ago

AI generated analysis - I do agree with the findings !!

Here’s a detailed analysis for the bug report, highlighting the payload differences between the two agent creation methods (Letta ADE vs. Astra App) and exploring why the behavior might differ when deleting agents from both environments.

Issue Summary

Observed Behavior:
- Deleting an agent from the Astra App deletes associated memory blocks (persona and human).
- Deleting the same agent from Letta ADE does not delete those blocks.
Expected Behavior:
- Deleting an agent should not remove associated persona and human blocks since these memory components can be reused across agents.

Analysis of API Payload Differences

1. Memory Structure

Letta ADE Payload:

The human and persona memory blocks are nested inside the "memory" object without explicit block IDs.

Human Block (ADE):

"human": {
  "name": "astra_human2",
  "label": "human",
  "value": "Name: astra_human2"
}

Persona Block (ADE):

"persona": {
  "name": "astra_persona2",
  "label": "persona",
  "value": "Act as ANNA ..."
}

Astra App Payload:

The human and persona blocks include explicit block IDs, suggesting that these blocks already exist and are linked rather than created during the agent creation.

Human Block (Astra App):

"human": {
  "id": "block-78513bd3-0162-4d8e-a73c-7f3b677396b6",
  "name": "astra_human2",
  "label": "human",
  "value": "Name: astra_human2"
}

Persona Block (Astra App):

"persona": {
  "id": "block-d1e02aa7-e929-4c5f-a4a2-98cd02ecf475",
  "name": "astra_persona2",
  "label": "persona",
  "value": "Act as ANNA ..."
}

Key Difference:

ADE: Memory blocks are created inline without explicit block IDs, implying that new blocks are generated for the agent.
Astra App: Memory blocks are reused by referencing existing block IDs. This suggests a relational dependency between agents and memory blocks.

2. Behavioral Hypothesis on Deletion

ADE API Call:
- Since memory blocks are created as part of the agent creation process (without pre-existing IDs), the agent deletion logic might be treating these blocks as internal to the agent and leaving them untouched during deletion.
Astra App API Call:
- The Astra App references existing memory blocks by their block IDs. Upon deletion, the backend logic might treat these blocks as part of the agent lifecycle and mistakenly delete them, even though they should be treated as shared resources.

3. Possible Root Cause

Deletion Logic Differences:
- ADE likely considers the memory blocks as newly created and does not remove them upon agent deletion.
- Astra App deletes the referenced blocks because the API logic treats the blocks as internal to the agent, even though they were reused.

Conclusion

The issue arises due to inconsistent memory block handling between ADE and Astra App. While ADE creates new memory blocks during agent creation, Astra App reuses existing blocks. This difference in behavior likely leads to the deletion of memory blocks when agents are removed through Astra App.

raolak commented 4 hours ago

Not passing the id to the persona and human block fixed the issue...but i think we should ignore id make sure that blocks are immutable and cant be deleted through agent deletion... can be deleted inly through agent and user template flows.

cpacker / MemGPT

BLOCKER - Deleting the agent is deleting Human and Persona blocks #1917

Bug Report: Inconsistent Agent Deletion Behavior

Issue Summary

Agent Creation - Letta ADE API Call

Agent Creation - Astra App API Call

AI generated analysis - I do agree with the findings !!

Issue Summary

Analysis of API Payload Differences

1. Memory Structure

Key Difference:

2. Behavioral Hypothesis on Deletion

3. Possible Root Cause

Conclusion