Per-Beam Local Context Variables in Beam

[NathanielEvry]

Why:
Users will be able to leverage per-beam local context variables within Beam, enabling the creation of multi-agentic, context-aware operations. This feature allows for more nuanced and dynamic interactions, particularly in use cases involving chain-of-thought-of-experts or mid-step retrieval-augmented generation (RAG).

Description:
The request is to introduce local context variables relevant to each Beam thread. These variables should be accessible within the Beam window, allowing each thread to operate with hyper-local context information. This would enhance the ability of multi-agent systems to maintain unique, context-sensitive operations per thread.

For example, users could use prompts like:

This is thread {{BEAM_THREAD_ID}} of {{BEAM_THREADS_COUNT}}, running {{THREAD_MODEL_NAME}}

Or:

Hello thread {{thread_number}} of {{threads_count}}, you have the unique name {{ @quick:'a tiny name related to the number {{thread_number}}' }}

The goal is to provide each Beam thread with a unique identity and context, enabling more sophisticated and individualized operations across multiple threads.

Per-Beam Local Context Variables:

• Introduce per-beam variables, allowing each Beam thread to have its own set of context-specific variables.
• Decide whether these variables should include user-defined variables or be limited to system/beam-specific variables.

Use Cases:

• Multi-Agentic Character Elements: Facilitate the creation of micro-agentic characters in a chorus, where each Beam thread operates with a unique context. Multi-persona interactions, fusions, more.
• Chain-of-Thought-of-Experts: Enable complex problem-solving scenarios by allowing different threads to contribute based on their specialized context.
• Mid-Step RAG: Enhance retrieval-augmented generation workflows by providing threads with context-aware retrieval capabilities.