People want tokenmaxx as the one local endpoint for every harness they run, not just the official apps. Today tokenmaxx install only rewrites the codex and claude configs.
Make install take a target: tokenmaxx install <openclaw|pi|hermes>, alongside today's codex and claude. Each target writes that harness's config to point at 127.0.0.1:8459/openai or /anthropic, with the same marked-block-and-restore approach we use now. tokenmaxx uninstall <target> puts the original back. doctor reports routing per target.
Where each config lives:
- OpenClaw: provider config with a base URL and API key. Since April it is API-key only for Claude.
- Pi: a provider entry with
baseUrl and apiKey.
- Hermes:
~/.hermes/config.yaml (model.base_url, plus api_mode: anthropic_messages or codex_responses), keys in ~/.hermes/.env.
Pairs with #10: harness traffic runs on API keys or extra usage, and Claude bills harness traffic to extra usage rather than the plan window. Routing a harness through tokenmaxx is how you actually see that spend.
Done when install and uninstall round-trip each harness's config safely, a harness pointed at tokenmaxx gets the same switching and metering as the official apps, and doctor tells the truth about every target.
Thanks @tedspare for the idea.
People want tokenmaxx as the one local endpoint for every harness they run, not just the official apps. Today
tokenmaxx installonly rewrites the codex and claude configs.Make install take a target:
tokenmaxx install <openclaw|pi|hermes>, alongside today's codex and claude. Each target writes that harness's config to point at127.0.0.1:8459/openaior/anthropic, with the same marked-block-and-restore approach we use now.tokenmaxx uninstall <target>puts the original back.doctorreports routing per target.Where each config lives:
baseUrlandapiKey.~/.hermes/config.yaml(model.base_url, plusapi_mode: anthropic_messagesorcodex_responses), keys in~/.hermes/.env.Pairs with #10: harness traffic runs on API keys or extra usage, and Claude bills harness traffic to extra usage rather than the plan window. Routing a harness through tokenmaxx is how you actually see that spend.
Done when install and uninstall round-trip each harness's config safely, a harness pointed at tokenmaxx gets the same switching and metering as the official apps, and doctor tells the truth about every target.
Thanks @tedspare for the idea.