feat: add support for coding plan providers (glm, kimi, minimax)

[ZerGo0]

This pr adds full support for GLM, Kimi, and MiniMax coding-plan.

Tested:

• GLM coding plan
• Kimi coding plan
• Minimax coding plan (I don't have coding plan anymore, but the implantation is based on their docs)

[nmb4]

how did you get the kimi coding plan to work? i tried to use it in codex last week i think and it errors saying only the claude code and kimi tools are allowed to call it's api.

[nmb4]

i just tried to look at the code to figure it out myself and saw that you also ran formatting, maybe you could tell your agent to split the commit into two, with one having all the formatting and a second that just contains the actual changes that implement the coding plan integrations?

[ZerGo0]

@nmb4 Good catch, I didn't test that, but I fixed it in: 285d12a

Openclaw uses pi for this stuff I think and they seem to use Anthropic/JS 0.73.0, but yea, the current user-agent works just fine.

[hotzen]

Thanks for the PR, will test MiniMax in a second

[hotzen]

works like a charm, thanks!

[llm]
minimax_key = "env:MINIMAX_API_KEY"

[defaults.routing]
channel = "minimax-coding/MiniMax-M2.5"
worker = "minimax-coding/MiniMax-M2.5"

[ricorna]

Hold off a min. I think i have a more elegant solution worked out. I'm testing it all currently.

this root issue is for many of these providers the API key is generic, you can use it to hit the paid API endpoint or the Coding subscription endpoint. no way to tell which is which by key itself though.

i'm normalizing the base urls used across the providers setup. This then allows you to always set a custom base URL in your provider settings, including in the UI for future proofing and maximum compatibility, and adding default options for the ones I can find and test. Z.ai and kimi code so far.

[ricorna]

Update: the alternative approach I mentioned is ready. Submitted as three PRs:

• feat: configurable base_url for all LLM providers #24 — Configurable base_url for all LLM providers: adds a generic base_url override to every provider (env var, TOML, and UI preset selector). This is the foundation — one provider, one key, URL is a config choice.
• feat: add Moonshot AI (Kimi) as native provider #25 — Moonshot AI (Kimi) provider: native Moonshot/Kimi support with Standard and Kimi Coding Plan endpoint presets.
• feat: add MiniMax as native provider #26 — MiniMax provider: native MiniMax support with International, China, Anthropic-compatible, and Coding Plan endpoint presets.

All three coding plans (Z.ai, Kimi, MiniMax) tested and verified with real subscriptions. The base_url approach covers everything your PR aimed to solve — plus proxies, Azure, self-hosted, regional endpoints, and any future tier variations — without multiplying provider entries.

Your research on the coding plan endpoints was genuinely helpful in getting this done — wouldn't have known the right URLs without your work here. Thanks @ZerGo0.

[RenderCoder]

I hope to be able to customize the model supplier URL.

[sneycampos]

Any updates for this PR or any chance to be merged? 🙏🏻

[thesammykins]

Hey! I've added the Z.AI coding plan and it's been merged #67