Connecting Kimi / GLM / a Custom OpenAI-Compatible Model

Step-by-step illustrated guide | For users who want more domestic/custom models | Reuses the DeepSeek setup template

Besides DeepSeek, Scholardo ships built-in presets for third-party models like Kimi and GLM; and if you have a custom service that's OpenAI-compatible, you can add it yourself. They connect exactly like DeepSeek — all in the merged single AI Agent tab, where you add a key + add the service. This guide is the "swap the model" version of P0-3 "Full DeepSeek Setup".

Reading P0-3 first makes this smoother — this guide only covers what's different.

---

One Shared Premise: They All Run on Top of Claude Code

Like DeepSeek, Kimi / GLM / custom OpenAI-compatible models all run on top of Claude Code — Claude Code acts as the "carrier." On an overseas network, the first time you launch one, Scholardo installs Claude Code for you automatically. You don't need to buy a Claude subscription for this.

⚠️ In mainland China: the one-click carrier install runs from claude.ai and is blocked (Kimi/GLM are the same as DeepSeek here). Install Claude Code via npm (domestic mirror) first — follow "Step 1" of Claude Code in Mainland China (skip its later env-var steps; the app injects each service's endpoint for you), then come back and configure below. Kimi (Moonshot) / GLM (Zhipu) are domestic services, so they run over a direct connection at runtime — no VPN needed.

---

Connecting Kimi or GLM (built-in presets)

The flow is identical to DeepSeek — you just pick a different preset:

1. First, sign up on the relevant platform, top up, and create an API key (the key is usually shown only once — copy it immediately):
   • Kimi: the Moonshot open platform.
   • GLM: the Zhipu AI open platform.
   • For sign-up/billing specifics, see each platform's current page.
2. In Scholardo Settings ▸ AI Agent:
   • In the AI providers section, click +, pick Kimi (or GLM), and paste the key.
   • In the CLI agents · Third-party section, click +; by now Kimi / GLM is ready (it shows (needs API key) when the key isn't filled in yet) — select it to add it.
3. Back in your project, open a new session → the service appears in "Choose an agent" → click it to launch.
   • If it shows "Needs Claude Code installed" + Install, click Install to let it install the carrier automatically.
   • When ready, the row shows a summary like <model name> · Claude Code.

---

Connecting a Custom OpenAI-Compatible Model

If your service provides an OpenAI-compatible API (a base URL + API key + model name), you can add it yourself:

1. Settings ▸ AI Agent ▸ AI providers section, click + → pick Add custom provider.
2. Fill in the service's base URL, API key, and model name (per its docs).
3. Go to the CLI agents · Third-party section, click + → Add custom service, wiring it into a runnable terminal agent.
4. Back in your project, open a new session → pick it in the chooser to launch.

🎯 Remember the shared mental model: the AI providers section holds the key; the CLI agents · Third-party section turns it into a runnable agent — both sections live in the same AI Agent tab.

---

Shot list (screenshot checklist)

Vendor UI (the sign-up / create-key pages of Kimi / GLM / a custom service platform) must be captured as-is — do not fabricate or redraw, and do not annotate specific price figures. Capture the Scholardo UI from the current release, using the merged single AI Agent tab. Redact keys in screenshots.

#	What to capture	UI source	Key callout
1	The create-API-key page on Kimi (Moonshot) / GLM (Zhipu)	Vendor UI (captured as-is, key redacted)	🔴 "the key is shown only once — copy immediately"
2	Settings ▸ AI Agent ▸ AI providers section + → pick Kimi / GLM → paste key	Scholardo (key redacted)	Circle the + and the key field
3	CLI agents · Third-party section + → Kimi / GLM ready (contrast with needs API key)	Scholardo	Circle "once the key is in, it lights up"
4	Custom: + → Add custom provider, fill in base URL/key/model	Scholardo	Circle the three inputs
5	Open a new session; the service appears in the chooser and launches	Scholardo	Side note: "runs on top of Claude Code; the carrier auto-installs on first launch"

---

Key callouts recap

1. Connects the same way as DeepSeek: AI providers holds the key + CLI agents · Third-party adds the service, in the same AI Agent tab.
2. Kimi / GLM are built-in presets; OpenAI-compatible services use Add custom provider / Add custom service.
3. They all run on top of Claude Code: the carrier auto-installs on first launch — no separate Claude subscription needed.
4. The key is stored only in your local Keychain — never uploaded.

---

Common pitfalls (see the full P0-6 · Install & Sign-in FAQ)

• A service shows needs API key in the chooser: go back to the AI providers section and fill in and save that service's key first.
• The terminal says it's not configured / can't read the key: return to the same section in Settings ▸ AI Agent, re-save the key, and open a new session (the third-party services' warnings match DeepSeek's — see P0-6).
• A custom service won't connect: double-check that the base URL / model name match the service's docs, and that it's genuinely an OpenAI-compatible API.

---

Data Privacy Disclosure

When you talk with Kimi / GLM / a custom service, your manuscript content is sent to the corresponding model provider for processing (Moonshot / Zhipu / whichever provider your custom setup points to). Your API key is stored only in your local Keychain and is never uploaded; but how the manuscript content you send is handled is governed by each provider's own privacy policy. If your research materials involve unpublished data, participant privacy, or confidentiality agreements, first confirm that sending them to a third-party AI service complies with your institution's requirements.