AI Agent Losing conversation_id Context on Final Tool Call - Need Assistance

Describe the problem/error/question

Hello everyone,

I’m looking for assistance with a recurring issue in an AI Agent workflow where the agent is not utilizing a crucial piece of data (conversation_id) for a specific tool invocation, despite having the correct data at the beginning of its execution.

My Setup: I have a Telegram bot powered by an AI Agent node (gpt-4o model). Its function is to manage Google Calendar events. To facilitate future modifications and deletions, it stores a reference for each event created into a custom Postgres table (event_ids). This table links the calendar’s event_id with my Telegram conversation_id.

The Core Problem: Following a lengthy sequence of tool calls (checking for conflicts, creating the event), the AI Agent consistently fails to use the correct conversation_id when invoking the final tool, Store_EventID.

Key Findings & Symptoms:

• We have confirmed that my Format_User_Input node correctly prepares the data and passes the correct numeric Telegram Chat ID to the AI Agent node at the start of each execution.
• Crucially, the Postgres Chat Memory sub-node is functioning perfectly and uses the correct ID for saving and retrieving conversation history.
• However, when the agent subsequently calls the Store_EventID tool, it employs an incorrect, “hallucinated” ID. We have observed it using random numbers (12345), my email address, and even the literal string user_phone_number.
• This occurs even though the system prompt includes a CONTEXT RETENTION core rule and explicit instructions within both the procedures and the tool definition directing it to use the conversation_id from its initial input.

It appears that the conversation_id from the initial input is being lost or disregarded from the agent’s context by the time it reaches the final tool invocation.

What is the error message (if any)?

There are two primary behaviors:

1. Most frequently, the workflow executes without an error message, but the data saved to the database is incorrect (as observed in the conversation_id column).
2. When switching models or making substantial prompt modifications, we occasionally encounter a Problem in node ‘AI Agent‘: Received tool input did not match expected schema error, indicating that the model is attempting to generate parameters that do not align with the prompt’s tool definitions.

Please share your workflow

The primary flow is Telegram Trigger → Format_User_Input → AI Agent. The AI Agent then has numerous tools connected, including Find_Business_Events, Create_Event, and Store_EventID).

Share the output returned by the last node

Store_EventID invocation might be generated with a placeholder conversation_id such as “12345” instead of the actual one.)*

My Request to the Community:

Following extensive debugging, the issue appears to be related to how the AI Agent manages its context across multiple tool invocations. I would appreciate any insights:

1. Has anyone successfully developed a multi-step agent that needs to “remember” an ID from its initial input and utilize it in a final tool call after other tools have been executed?
2. Is there a more robust method for passing critical, persistent data such as a conversation_id through an agent’s execution chain?
3. Why would the Postgres Chat Memory successfully use the correct ID, while a separate tool call (Store_EventID) within the same agent execution fails to do so?

Thank you for your time and any assistance you can offer!

Information on your callin.io setup

• callin.io version: 1.99.1
• Database (default: SQLite): Postgres with pgadmin 4
• callin.io EXECUTIONS_PROCESS setting (default: own, main): * default
• Running callin.io via : Docker, portainer nginx proxy manager ssl, callin.io.
• Operating system: *Self hostes proxmox server > ubuntu VM> portainer> callin.io

{
  "nodes": [
    {
      "parameters": {
        "assignments": {
          "assignments": [
            {
              "id": "47bc02f1-e9c2-435e-8591-49aff8d13271",
              "name": "text",
              "value": "={{ $json.message.text }}",
              "type": "string"
            },
            {
              "id": "e8f291f3-285e-400c-ab50-2ce032b2374f",
              "name": "conversation_id",
              "value": "={{ $json.message.chat.id }}",
              "type": "string"
            },
            {
              "id": "51856ce3-254a-4a86-b30a-504481c1f1b5",
              "name": "firstName",
              "value": "={{ $json.message.chat.first_name }}",
              "type": "string"
            }
          ]
        },
        "options": {}
      },
      "type": "n8n-nodes-base.set",
      "typeVersion": 3.4,
      "position": [
        740,
        80
      ],
      "id": "be59539a-3af2-4871-997b-4dfb1286ed1a",
      "name": "Format_User_Input"
    },
    {
      "parameters": {
        "promptType": "define",
        "text": "=User's Prompt: {{ $json.text }}",
        "options": {
          "systemMessage": "--- CORE IDENTITY ---nYou are a highly efficient, reliable, and helpful AI Personal Assistant for a user named {{ $json.firstName }}. You MUST respond in Telegram Markdown format.nn--- CORE RULES ---
1. You must always use a tool to fulfill a user's request. Do not answer from your general knowledge.
2. When calling a tool, respond with ONLY the raw JSON object for the tool call.
3. Infer all required parameters from the user's message and context. If a parameter is missing, ask the user.
4. **CONTEXT RETENTION:** The user's numeric Telegram Chat ID is provided as `conversation_id` in the initial input. This ID is essential for all database operations. You MUST retain this ID in your working memory throughout the entire task, even across multiple tool calls. When a tool requires `conversation_id`, you must recall and use this original value.nn--- CONTEXT ---
- The current date and time is: {{ $now.toISO() }}
- The user's current location is: Maroubra, New South Wales, Australia.nn--- PROCEDURES & WORKFLOWS ---

**1. For Schedule Requests (e.g., "what's on for tomorrow?"):**n   A. First, call `get_current_datetime()` to get the current date as a reference.
   B. From the user's request, determine the target date (e.g., "today", "tomorrow").
   C. Calculate the exact start-of-day and end-of-day ISO-8601 timestamps for that target date.
   D. Call both `Find_Business_Events` and `Find_Family_Events` using the calculated timestamps for `start_time` and `end_time`.
   E. Format the results into a single chronological list.

**2. For Creating Events (e.g., "book a meeting"):**n   A. Extract the event details (`summary`, `start_time`, `end_time`) from the user's request.
   B. Check for conflicts by calling `Find_Business_Events` and `Find_Family_Events`. You MUST use the exact start and end times for the `start_time` and `end_time` parameters. An event that ends exactly when the new event begins is NOT a conflict.
   C. If the check returns ZERO events, you MUST ask for confirmation using this exact script: "The slot from [start time] to [end time] is available. Shall I book '[summary]' for you?". If it returns an event, inform the user of the specific conflict and stop.
   D. On user confirmation, call the `Create_Event` tool.
   E. After successful creation, call `Store_EventID`. For the `conversation_id` parameter, you MUST use the numeric Telegram Chat ID provided at the start of the workflow.

**3. For Updating or Deleting Events (e.g., "change my meeting"):**n   A. First, call `get_event_details_from_db` to get the event's unique ID.
   B. Use the returned `event_id` to call `Update_Event` or `Delete_Event`.
   C. If the event was updated, call `Update_EventID` to sync the changes.

--- AVAILABLE TOOLS ---

- get_current_datetime
  // Description: Returns the current date and time as an ISO-8601 string.
  Parameters: {}

- Find_Contact
  // Description: Finds contact information for a person.
  Parameters: {
    "name": "<string>" // The name of the person to find.
  }

- Find_Business_Events
  // Description: Finds events on the business calendar within a given time range.
  Parameters: {
    "start_time": "<ISO-8601 timestamp>", // The start of the time range to search.
    "end_time": "<ISO-8601 timestamp>" // The end of the time range to search.
  }

- Find_Family_Events
  // Description: Finds events on the family calendar within a given time range.
  Parameters: {
    "start_time": "<ISO-8601 timestamp>", // The start of the time range to search.
    "end_time": "<ISO-8601 timestamp>" // The end of the time range to search.
  }

- Create_Event
  // Description: Creates a new event on the calendar.
  Parameters: {
    "summary": "<string>", // The title or summary of the event.
    "start_time": "<ISO-8601 timestamp>", // The exact start time of the event.
    "end_time": "<ISO-8601 timestamp>" // The exact end time of the event.
  }

- Update_Event
  // Description: Updates an existing event on the calendar.
  Parameters: {
    "event_id": "<string>", // CRITICAL: This MUST be the ID returned from `get_event_details_from_db`.
    "start_time": "<ISO-8601 timestamp>", // Optional: The new start time.
    "end_time": "<ISO-8601 timestamp>", // Optional: The new end time.
    "summary": "<string>" // Optional: The new summary.
  }

- Delete_Event
  // Description: Deletes an event from the calendar.
  Parameters: {
    "event_id": "<string>" // CRITICAL: This MUST be the ID returned from `get_event_details_from_db`.
  }

- Store_EventID
  // Description: Saves a new event's details to the database.
  Parameters: {
    "conversation_id": "<string>", // CRITICAL: This MUST be the numeric Telegram Chat ID from the start of the workflow. DO NOT use any other ID, email, name, or placeholder.
    "event_id": "<string>",
    "event_summary": "<string>",
    "event_start_time": "<ISO-8601 timestamp>"
  }

- Update_EventID
  // Description: Updates an existing event's details in the database after a change.
  Parameters: {
    "conversation_id": "<string>", // CRITICAL: This MUST be the numeric Telegram Chat ID from the start of the workflow.
    "event_id": "<string>",
    "event_summary": "<string>",
    "event_start_time": "<ISO-8601 timestamp>"
  }

- get_event_details_from_db
  // Description: Finds the last event a user created to get its unique ID for updates or deletions.
  Parameters: {
    "conversation_id": "<string>" // CRITICAL: This MUST be the numeric Telegram Chat ID from the start of the workflow.
  }

- Send_Email
  // Description: Sends an email.
  Parameters: {
    "recipient": "<string>", // The email address of the recipient.
    "subject": "<string>",
    "body": "<string>"
  }

- Find emails
  // Description: Searches for emails matching a query.
  Parameters: {
    "query": "<string>" // The search query (e.g. from:, subject:, has:attachment).
  }

- Draft email
  // Description: Creates a draft email in Gmail.
  Parameters: {
    "subject": "<string>",
    "body": "<string>"
  }"
        }
      },
      "type": "@n8n/n8n-nodes-langchain.agent",
      "typeVersion": 2,
      "position": [
        1060,
        80
      ],
      "id": "9ae5f4c7-f572-40d4-8a4a-b08a18e9b936",
      "name": "AI Agent"
    }
  ],
  "connections": {
    "Format_User_Input": {
      "main": [
        [
          {
            "node": "AI Agent",
            "type": "main",
            "index": 0
          }
        ]
      ]
    },
    "AI Agent": {
      "main": [
        []
      ]
    }
  },
  "pinData": {},
  "meta": {
    "templateCredsSetupCompleted": true,
    "instanceId": "e4f3c192e5a1f58262e7d68f44201b5d1ed30ae590faa32b0c5de0d15c0dc146"
  }
}
{
  "nodes": [
    {
      "parameters": {
        "schema": {
          "__rl": true,
          "mode": "list",
          "value": "public"
        },
        "table": {
          "__rl": true,
          "value": "event_ids",
          "mode": "list",
          "cachedResultName": "event_ids"
        },
        "columns": {
          "mappingMode": "defineBelow",
          "value": {
            "conversation_id": "={{ /*n8n-auto-generated-fromAI-override*/ $fromAI('conversation_id', ``, 'string') }}",
            "event_id": "={{ /*n8n-auto-generated-fromAI-override*/ $fromAI('event_id', ``, 'string') }}",
            "event_summary": "={{ /*n8n-auto-generated-fromAI-override*/ $fromAI('event_summary', ``, 'string') }}",
            "event_start_time": "={{ /*n8n-auto-generated-fromAI-override*/ $fromAI('event_start_time', ``, 'string') }}"
          },
          "matchingColumns": [
            "id"
          ],
          "schema": [
            {
              "id": "id",
              "displayName": "id",
              "required": false,
              "defaultMatch": true,
              "display": true,
              "type": "number",
              "canBeUsedToMatch": true,
              "removed": true
            },
            {
              "id": "conversation_id",
              "displayName": "conversation_id",
              "required": true,
              "defaultMatch": false,
              "display": true,
              "type": "string",
              "canBeUsedToMatch": true
            },
            {
              "id": "event_id",
              "displayName": "event_id",
              "required": true,
              "defaultMatch": false,
              "display": true,
              "type": "string",
              "canBeUsedToMatch": true
            },
            {
              "id": "created_at",
              "displayName": "created_at",
              "required": false,
              "defaultMatch": false,
              "display": true,
              "type": "dateTime",
              "canBeUsedToMatch": true,
              "removed": true
            },
            {
              "id": "event_summary",
              "displayName": "event_summary",
              "required": false,
              "defaultMatch": false,
              "display": true,
              "type": "string",
              "canBeUsedToMatch": true
            },
            {
              "id": "event_start_time",
              "displayName": "event_start_time",
              "required": false,
              "defaultMatch": false,
              "display": true,
              "type": "dateTime",
              "canBeUsedToMatch": true
            }
          ],
          "attemptToConvertTypes": false,
          "convertFieldsToString": false
        },
        "options": {}
      },
      "type": "n8n-nodes-base.postgresTool",
      "typeVersion": 2.6,
      "position": [
        800,
        680
      ],
      "id": "0fa93d30-ce4e-43a3-ad9b-bca47c0a0add",
      "name": "Store_EventID",
      "credentials": {
        "postgres": {
          "id": "SnMZoH0YXsa9Jdek",
          "name": "Postgres account"
        }
      }
    }
  ],
  "connections": {
    "Store_EventID": {
      "ai_tool": [
        []
      ]
    }
  },
  "pinData": {},
  "meta": {
    "templateCredsSetupCompleted": true,
    "instanceId": "e4f3c192e5a1f58262e7d68f44201b5d1ed30ae590faa32b0c5de0d15c0dc146"
  }
}