Token Economics and Provider Choice
Choose the LLM provider and model from configured API keys — token consumption visible through a fuel-gauge with Healthy, Caution, Warning, and Trimming states; compaction events are disclosed, not silent.
Token economics — how much context the AI model can use, how much it costs, and what happens when the conversation grows large — are usually invisible to users of AI-integrated products. The conversation just slows down, or becomes less coherent, or quietly starts forgetting earlier context, and the user has no way to understand why. ProjectXL takes a different approach: token consumption and conversation state are visible to the user at all times.
Provider and Model Transparency
Users choose which LLM provider and model ProjectXL uses for AI assistance by configuring their API keys and selecting from the available options. The selection is explicit — there is no automatic provider switching or silent fallback to a different model when budget is constrained. If the user needs to change providers or models, they do so deliberately. The provider and model currently in use are always visible.
Conversation State Management
As an AI conversation grows, the token footprint grows with it. When the conversation approaches the model's context limit, some older content must be dropped to make room for new turns. Most applications handle this silently — content disappears from the context window without any indication to the user. ProjectXL surfaces this as a deliberate, visible event. Before dropping older turns, the system compacts them into a structured session-state summary that preserves the durable meaning of the prior conversation: the task context, design decisions, user constraints, and current implementation state. The compaction notice describes what was preserved and what was released, so the user can judge whether the summary captures what matters before the conversation continues.