TokenScope is the planning desk for people shipping language model features. Four instruments, zero jargon.
The Model Planner compares cost, latency, and throughput across five model tiers from an editable pricing table, so every number is yours, not ours. Memory Fit calculates weights and KV cache for your context length at each quantization level, and shows the arithmetic line by line. The Batch Simulator is a deterministic queueing model: move arrival rate and batch size and watch p fifty and p ninety five respond, with a full data table as the accessible alternative to the chart. The Route Builder composes escalation rules, cheap model first, premium only past a confidence threshold, and replays them against a synthetic request log to show the savings.
A seven step guided tour walks a first time builder through every panel, because this hackathon welcomes beginners and so does the tool. One hundred unit tests cover the pricing math, memory arithmetic, queueing simulation, and route replay. No live API is called; the Nebius integration point is documented and marked in the code.
Log in or sign up for Devpost to join the conversation.