chimera.providers.cost_tracker

Two modules cooperate to track LLM spend:

Pricing (`chimera.providers.cost`)

from chimera.providers.cost import register_model_cost, get_model_cost

Built-in pricing for every model the catalog ships (GLM, Claude, DeepSeek, Kimi, Qwen3, GPT-OSS, …). Register your own:

register_model_cost("acme/llm", 0.50, 1.50)  # $/Mtok input, output

Returns input / output USD per million tokens.

from chimera.providers.cost_tracker import CostTracker

CostTracker records granular per-step token counts (cache hits, reasoning tokens, regular input/output) and rolls them into a session total.

Method	Purpose
`record_step(usage)`	Append a step’s token usage.
`total_tokens()`	Sum across all steps.
`total_cost()`	USD across all steps.
`breakdown()`	Per-step rows with cache / reasoning / regular splits.

Wired into the loop via LoopConfig.cost_tracker= and surfaced in the REPL via /cost.