prompt truncated

### What is the issue? Hello, # Loss of Context When Using Tools in a Conversation with Ollama ## Description When using Ollama to verify a user's access rights, I noticed that the **generated prompt** after the initial processing is truncated. The issue manifests as the loss of initial context or information (e.g., details about available tools or system instructions). This prevents the model or agents from effectively utilizing the defined tools, reducing the relevance of the responses. Despite sending a complete initial message with clear instructions and defined tools, the **generated prompt** no longer reflects this information and is reduced to the user's simple request. --- ## Steps to Reproduce the Issue 1. Send an initial complete message with: - A `system` role specifying the context. - A list of available tools in the `tools` field. - A clear user request. 2. Observe that the system correctly handles the first message. 3. Check the **generated prompt** for tools in the model's response. 4. Observation: the **generated prompt** is incomplete and does not retain the initial context. --- ## Expected Behavior The **generated prompt** for tools should include: 1. The initial instructions (defined in the `system` role). 2. Details about the available tools. 3. A reformulation of the user request within the defined context. --- ## Observed Behavior The **generated prompt** is simplified and only contains: - The user's direct request. - Missing initial contextual instructions and defined tools. --- ## Examples of Messages and Prompts ### Initial Message Sent: <details> { "model": "mistral-large:123b", "messages": [ { "role": "system", "content": " \nVous êtes un agent LLM chargé de vérifier les droits d'accès d'un utilisateur. Votre tâche consiste à utiliser les outils et fonctions disponibles pour répondre aux questions relatives aux droits d'accès. \nLa demande n'est pas forcément direc…