track LLM CLI usage windows with ambient feedback: keyboard brightness, voice readouts, and an HTTP /status API for consumers like utop.

darker keyboard = less remaining usage; resets to full brightness when the tracked window resets.

providers

• claude: polls anthropic oauth usage windows (five-hour + weekly windows from /limits)
• codex: polls codex/chatgpt usage API (https://chatgpt.com/backend-api/wham/usage)
• codex_logs: reads local codex rollout logs from ~/.codex/sessions and uses the same /status rate-limit windows

configure provider and tracked window in config.yaml.

tasks

todo

ongoing

active

passive

• account support: ses_2ac4222c4ffeb3hlQz0LSMF5cs
• optimization/utop: ses_2aa243ba1ffe7gZUNwAfLlKrL5

waiting

• Codex support via API - ses_344c1d07cffeBslyiLe6CdRXbA
  • initial API implementation
  • monitor for stability

done

• rename brightness-monitor to llm-usage; migrate setuptools+uv to poetry; fix mac pyobjc propagation in prism-shared-python - ses_2a4924e52ffeGhOPXjqH008YDR

cancelled