Strategies run in order · the first to answer wins
Claude, Codex, Gemini, Cursor, Copilot, DeepSeek, Kimi, z.ai, Qwen, Cline, Ollama Cloud and more.
Providers
How Runic learns what each provider has used. This is the summary; the per-provider detail is docs/providers.md and docs/<provider>.md in the repo, which is the source of truth.
Runic tries a provider’s strategies in order and uses the first that answers. The source label shown in the CLI and in Settings → Providers is one of openai-web, web, oauth, api, local, openTelemetry, or a provider CLI label such as codex-cli or claude.
Subscription tools
| Provider | Strategies, in order | Notes |
|---|---|---|
| Codex | Web dashboard → CLI RPC → CLI PTY /status | Today’s cost comes from ~/.codex/sessions; older days from the relay. |
| Claude | OAuth API → web API → CLI PTY | See Claude. |
| Gemini | OAuth via the Gemini CLI’s credentials | |
| Cursor | Browser cookies → stored WebKit session | On-demand spend shown without a gauge when the plan is unlimited. |
| Droid / Factory | Cookies → stored tokens → local storage → WorkOS | |
| Copilot | Device-flow token → copilot_internal API | |
| Antigravity | Local LSP / HTTP probe | |
| Muse Code (Meta) | Local muse CLI /usage | An explicit API mode uses a pay-as-you-go key. |
| opencode | Local session logs | History only, no live gauge. |
Keys and balances
| Provider | Source | What you see |
|---|---|---|
| DeepSeek, Kimi, StepFun, OpenRouter, Vercel AI | API key → balance API | Balance, plus spend derived from readings taken every 20 minutes; a “From logs” estimate where the provider can be routed through a coding tool. |
| z.ai, MiniMax, Qwen | API key or web session → quota API | Quota windows. Qwen token and coding plans report usage from local coding-tool logs. |
| Cline | API key → api.cline.bot | Credit balance and exact spend today and this month. |
| Ollama Cloud | API key → ollama.com/api/usage | Monthly credits, or session and weekly limits. Undocumented, parsed defensively. |
| TypeSafe | API key → /v1/models | Key health only; the published API reports no usage. |
| Local LLM | Runtime probe (Ollama, LM Studio, vLLM, llama.cpp, Open WebUI) | Usage comes from OpenTelemetry GenAI or local logs when configured. |
Two regions
China and international are separate providers, not a setting: Kimi, z.ai, MiniMax, Qwen and StepFun each have a CN slot. The tab strip shows one tab per brand and the card stacks a section per configured region, with an “Add China account” or “Add International account” row for the other side.
Credentials
Keys live in your login Keychain. Runic-signed builds are the only binaries that touch it; the CLI and unsigned test builds read environment variables instead. Nothing is sent anywhere except the provider’s own API.