ZCode Model Setup

⬅️ Coming from

Step 2 · Enable Core Capabilities set up permissions and capabilities → this step connects a model: prefer the metered DeepSeek or GLM API, with the built-in plan as a lightweight fallback.

🌟 Goal: connect the metered DeepSeek or GLM API and pass the connection test — that completes Step 3. About 15 minutes.

🎯 Why the Metered API Beats the Plan

Plan-style options (BigModel built-in plan / GLM Coding Lite, etc.) have two hard limits:

• 5-hour quota window: when the quota runs out, you wait for the next 5-hour refresh — getting cut off mid-task is the worst;
• Weekly usage cap: heavy weeks hit the cap early, leaving you idle for days.

DeepSeek / GLM metered APIs have neither: pay as you go, charged per token — heavy work won't get cut off. For everyday outreach emails and document work, spending is usually low and controllable.

⚡ Quick Answer

Recommended: DeepSeek API or GLM API (metered, no 5-hour / weekly limits)
Zero-cost fallback: BigModel built-in plan + GLM-5.3-Flash (quota windows apply, fine for light trials)

• Recommended: connect the DeepSeek API (cheap and stable) or the GLM API (the account you signed into in Step 1 is already the Zhipu open platform — getting a Key is the shortest path);
• Zero-cost trial: start with the built-in plan to get a feel, switch to the API before heavy use;
• Connecting both is fine — when you install the AI agent team later, different roles can use different models.

✅ Recommended Route: Connect a Metered API

Both providers have dedicated guides with full steps, screenshots, and pitfall tables:

• 📚 Connect GLM API to ZCode (Pay-as-you-go) — shortest path: you get the Key from the Zhipu open platform (the same site you signed into in Step 1): register → top up → create Key → fill into ZCode → test;
• 📚 Connect DeepSeek to ZCode: Complete Guide — register at platform.deepseek.com → top up → create Key → fill into ZCode → test.

The fields to fill are the same for every provider:

Field	Notes
Provider name	Anything you recognize, e.g. DeepSeek
API format	Pick the one available in your ZCode dropdown
Base URL	Must be the same set as the API format; copy exactly
API Key	Create in the provider console; treat like a password
Model ID	Per the ZCode list or the provider console

Two easiest traps

1. API format and Base URL must be the same set — never pair provider A's address with provider B's format.
2. Copy the address exactly — don't add /v1 or a trailing slash yourself; each provider differs (see the cheat sheet below).

Click Test after filling — it only counts when you see a green "connected" message.

If the test fails, check two things first: ① do the format and address belong to the same set; ② was the model ID copied exactly from the list or console. If it still fails, follow that provider's guide step by step instead of blindly retrying combinations.

🔑 Treat the API Key like a password: fill it only into ZCode; never paste it into chats, logs, screenshots, tutorials, or public repos; mask it fully before screenshotting. On unusual spend, disable and recreate the key at the provider console immediately.

Once connected, open the model menu below the new-task input and pick the API's main model as your everyday default (e.g. DeepSeek's deepseek-chat, or GLM's GLM-5.3-Flash).

🟡 Fallback: Built-in Plan

If you'd rather not top up an API right away, start with the built-in plan: click Upgrade in the lower-left account menu, enable a plan (free or paid, whichever is currently offered), and pick GLM-5.3-Flash as the everyday model.

The trade-off is exactly the two limits above: a 5-hour quota window plus a weekly usage cap. Fine for a light trial; as the main engine for outreach runs, you'll keep hitting the quota wall.

Provider Cheat Sheet

The table lists Chat Completions format entries (paths verified by unauthenticated probe on Sep 8, 2026 — availability for your account or region is not guaranteed). When unsure, prefer Chat Completions; follow the provider docs if they require another format.

Provider	Base URL	Guide / official entry
DeepSeek	https://api.deepseek.com	📚 DeepSeek complete guide · Key page
Zhipu GLM (BigModel)	https://open.bigmodel.cn/api/paas/v4	📚 GLM API pay-as-you-go · Key page
Kimi (Moonshot)	https://api.moonshot.cn/v1	Key page · Model list
Alibaba Bailian (Qwen)	https://dashscope.aliyuncs.com/compatible-mode/v1	Bailian console
SiliconFlow	https://api.siliconflow.cn/v1	Key page · Model square

Multiple Zhipu-related entries — pick the right one

ZCode may show the built-in plan, a built-in API-key provider (both Anthropic format), and your own metered provider. Only the one you created yourself is the pay-as-you-go route; see the GLM guide.

Want screenshots?

This section is a field cheat sheet only. For the full screenshot walkthroughs:

• 📚 Connect DeepSeek to ZCode: Complete Guide
• 📚 Connect GLM API to ZCode (Pay-as-you-go)

Overseas models (e.g. Google Gemini) are out of scope here; follow the current Google AI Studio docs when needed.

📋 Common Mistakes

Pitfall	Consequence	How to avoid
🔴 Pasting the Key somewhere public	Strangers spend your money	Fill it only into ZCode; mask screenshots; disable and recreate on unusual spend
🔴 Skipping the Test button	You think it's configured but it isn't connected	Require a green "connected" before moving on
🟡 Format and address mismatch	Connection fails	Use the combination from the same row; copy addresses exactly
🟡 Model ID typed by hand or outdated	"Model not found" error	Copy from the ZCode list or the provider console
🟡 Using the built-in plan as the main engine for heavy work	The 5-hour / weekly limits keep cutting you off	Switch to DeepSeek / GLM metered APIs for heavy use

✅ Completion Check

Metered API connected (recommended route)

• The provider shows "enabled";
• API format and Base URL come from the same set;
• Model ID copied from the ZCode list or provider console;
• Test shows "connected" and the plain-text test passes;
• The API model is selected as the everyday default in the model menu;
• If you need image capabilities, confirmed per official docs and tested with a plain file.

Built-in plan for now (lightweight fallback)

• A usable BigModel plan is selected;
• GLM-5.3-Flash is selected for new tasks;
• You know the quota has a 5-hour window and a weekly cap, and will switch to the API before heavy use.

📚 Further Reading

• 📚 Step 2 · Enable Core Capabilities: Full Access, computer control, browser control, and more
• 📚 Step 4 · Install the ZCode AI Agent Team: different models per role
• 📚 Connect DeepSeek to ZCode: Complete Guide: sign-up, Key, fields, troubleshooting
• 📚 Connect GLM API to ZCode (Pay-as-you-go): no plan, charged from API balance

🚀 Next

📄 Step 4 · Install the ZCode AI Agent Team — with the metered API connected, DeepSeek + GLM give different roles real division of labor.