Command line version (Linux/server)
wink-run/tokenbank · skills
Token Bank — the local LLM gateway that sits between your AI agents and every provider.
★ 91GitHub stars
11Forks
2026-08-15Last updated
JavaScriptLanguage
Apache-2.0License
Key features
- Token Bank is more than an API proxy — it brings Claude Code, Codex, Cursor, WorkBuddy, Kimi Code, OpenClaw, and other mainstream agents under one local gateway.
- Claude Code / Codex CLI / OpenCode / Hermes / Kimi Code: CLI shim: injects BASE_URL (and related) env vars — no command changes
- Agents keep their native model names (claude-sonnet-4-6, gpt-5, …).
- Usage is traced whether or not traffic goes through the gateway
- No agent-side changes required for usage tracing, third-party model switching, and smart routing.
Requirements
- Prefer Gateway → Track for one-click onboarding — no manual env vars.
- Pick a route and the agent keeps its native model names while the gateway transparently forwards to your chosen provider.
Install command
dsh plugin --profile web add github:wink-run/tokenbank