🔖 Feature description

Implement a Jinja2-based template rendering system that allows prompts to dynamically inject context at runtime using organized namespaces.

🎤 Why is this feature needed ?

Currently, prompts are static strings with basic placeholder replacement ({summaries}, {query}). This severely limits dynamic context injection - you can't pass user information, system metadata, or pre-computed data into prompts. The system has no way to know the current date/time, request ID, or who's asking the question. Additionally, when tools need to be used (like fetching memory), the LLM must make a decision and call them after generating initial text, requiring a second LLM call just to use that context - a wasteful round trip.

✌️ How do you aim to achieve this?

A proper template system with namespaces enables:

• Dynamic Context at Runtime: Inject user info ({{ passthrough.user_name }}), system data ({{ system.date }}), RAG results ({{ source.documents }}), and pre-fetched tool data ({{ tools.memory.fetch }}) directly into prompts before the LLM sees them
• Single LLM Call: Pre-execute tools and render all context upfront, eliminating the need for the LLM to request data - it already has it
• Organized Variable Injection: 4 namespaces (system, passthrough, source, tools) prevent naming conflicts and make prompts readable
• Better Widget Integration: Web requests can pass custom parameters (user context, auth tokens, custom fields) that flow directly into prompts - enabling truly dynamic chatbot behavior

🔄️ Additional Information

No response

👀 Have you spent some time to check if this feature request has been raised before?

• I checked and didn't find similar issue

Are you willing to submit PR?

Yes I am willing to submit a PR!