Skip to content
ORACLE scry
// adapters-snapshot

Adapters & snapshot

token-oracle is provider-agnostic at the core: adapters emit neutral usage events; the engine projects caps; consumers read forecast.json.

Sources (high level)

  • Claude Code — local session transcripts / usage logs on disk
  • Grok Build — local Grok session trails
  • Optional live overlay — browser-verified numbers (opt-in; never invents)

Snapshot contract

Statuslines, tmux, herald, and judges should consume the snapshot file — not scrape the TUI.

Typical location (see config): under the oracle data dir as forecast.json (schema version 1).

Properties that matter to consumers:

  • per-window used / cap / projected %
  • ETA / time-left when known
  • honest states when data is missing (never fake green)

Deep adapter interface notes for integrators live in the repo ADAPTERS.md (source + consumer interfaces). Public install path stays offline-first; live is optional.