Summary: Allow users to switch between different AI models and Pickaxe agents mid-conversation while maintaining a single, continuous, shared chat history.
The Problem: Currently, interacting with multiple specialized agents forces users into isolated chat histories for each one. Pickaxe promotes a “true multi-model advantage in a single workspace”, but that workspace remains fragmented. It forces copy-pasting of context, or connecting to other Pickaxe agents as Actions, which requires at least two extra model queries, acting as a major bottleneck and token consumer for operations that require cross-agent collaboration. In addition, Credits end users might want to use a cheaper model for a majority of requests and a larger model for more complicated requests.
For example, my users might start a conversation on the Chat endpoint, and decide that they need to run a simulation on the Simulation endpoint based upon the conversation, and want to use a larger model for that simulation. In my LibreChat solution, they click on the model selector and choose the Simulation agent and select a model, and continue the conversation. In Pickaxe, they would either have to start a new conversation with the other agent and copy-page the context, or the model would have to do a tool call to the other agent and then evaluate the response, which is two additional model queries per main agent model query, as opposed to switching to the Simulation agent and selecting a model and continuing the conversation with the current context in one model query. These users will undoubedly want to continue the conversation with the Simulation agent, which means running two extra model queries for every main agent model query, for a total of three model queries per message, and also requires trusting that the Chat agent will always decide to call the Simulation agent Action/tool, and directives that cover agent selection are not always applied in every situation. This pattern holds across all of my agents, including Chat, Scholar, Simulation, News Edition, Cowboy Coder, and Diamond Tech Support Agent.
The Solution: Implement a custom model and agent list and a model selection dropdown within the chat interface to change the active model and/or agent on the fly. The conversation history acts as a shared context pool, allowing the new agent to simply read the existing thread and continue the conversation. This has worked in practice in my LibreChat and Open WebUI solutions, and model switching works in practice in the Gemini app. In the admin UI, instead of selecting one model for the agent, multiple models can be selected and the order in which they appear can be set, and in the dropdown, the relative cost, currently denoted by “$$$” in the admin UI, can also appear for monetized agents that are set to use Credits instead of Message quotas. This is how it works in my custom LibreChat model selection solution.
The Impact: Adding this functionality mirrors the flexibility found in other chat interfaces, like LibreChat, Open WebUI, and the Gemini app, and allows model switching and complex, multi-agent orchestrations to occur naturally within a single user session, eliminates the current admin multi-agent action configuration complexity using Actions, saves a large number of tokens, and makes chat histories easier to process by eliminating multiple tool call and result entries for processing agent queries.
This feature would need to be available to Pro users so that I would be able to use it. For reference, I was able to add this feature to LibreChat and my custom FastAPI endpoint within a single day, not including the required days of testing afterwards.
Thank you very much for taking community feature requests seriously!
