Connect DeepSeek to ZCode: Complete Guide

⬅️ Coming from · Step 3 · Configure Plans and Models already set up a built-in plan and an everyday model → this guide adds a DeepSeek model on top.

💡 What you'll finish here: confirm balance (top up a little if needed) → get an API key → connect and test it in ZCode. About 15 minutes.

🗺️ The Flow

🎯 When to Use It

ZCode's built-in GLM Coding plan has a quota: Zhipu documents a dual cap of "every 5 hours and every week," and once it runs out you wait for the next window.

If you often work in long stretches (long tasks, organizing a batch of files, iterating on the same thing), you may hit the quota and have to stop.

DeepSeek bills differently: it charges your balance by actual usage, with no plan-style 5-hour/weekly quota pool, so it suits long continuous work.

⚠️ That does not mean DeepSeek has no limits. It is still bound by balance, concurrency caps, risk control, and service rules — follow the official pricing page and your own console.

🤔 DeepSeek or GLM?

They do not conflict — keep using the built-in plan and add DeepSeek as needed. The difference is billing and quota mechanics (not which is better):

	Built-in GLM Coding plan	DeepSeek
Billing	Buy a plan, spend its credit allowance	Balance charged by actual usage
Quota mechanics	Vendor documents a per-5-hour + per-week dual cap; when used up, wait for the next 5-hour window	No plan-style time-window quota pool; bound by balance and concurrency caps
Better for	Everyday, spread-out work	Long continuous tasks, batch processing
Extra setup	None, it comes with ZCode	You apply for a key and top up

💡 In short: use the built-in GLM for everyday work, switch to DeepSeek when you need to run for a long stretch. Each covers a different situation.

📚 Also: GLM can likewise be used without a plan, charged from API balance — see Connect GLM API to ZCode (Pay-as-you-go).

💰 What It Costs

DeepSeek is pay-as-you-go. If the account already has usable balance, you can skip this step; for a new account, start with a small top-up and add more later.

• Top up: 👉 DeepSeek top-up
• Pricing: 👉 DeepSeek pricing page

About usage

How long it lasts depends on how many tasks you run, what you send, and the current price. This guide promises no specific duration or cost — check the usage log in your console. Top-up only solves balance; it does not fix a wrong model ID, a format mismatch, or a permission problem.

🔑 Get an API Key

1. Open API Keys

Open the 👉 DeepSeek platform, sign in, then go to the 👉 API Keys page.

2. Create and copy it

1. Click Create API key at the top right;
2. Give it any name you'll recognize (for example zcode) and click Create;
3. In the dialog that appears, click Copy immediately — the key is shown in full only once.

An API key is a password

• Put it only into ZCode; never paste it into chats, screenshots, tutorials, issues, or public repositories;
• Mask it fully before taking any screenshot;
• Never upload or share ZCode's config file (such as config.json);
• If usage looks abnormal, disable the old key and create a new one in the console right away.

⚙️ Connect It in ZCode

1. Open model settings

Path:

Gear icon (lower left) → Settings → Model Settings

If DeepSeek is already listed, click it; otherwise click Add Provider.

2. Fill in the connection fields

Field	What to enter
Provider name	DeepSeek (anything you recognize)
API format	Responses (/responses)
Base URL	https://api.deepseek.com
API key	The key you copied above
Model ID	deepseek-flash (follow your ZCode list)

Input types are not part of this step — the next section covers them.

3. Which format

DeepSeek offers two formats with the same Base URL:

• This guide follows what we tested: Responses (/responses);
• If your ZCode dropdown only lists Chat Completions, use it, and verify capability against the same model and the official docs (the two formats are not guaranteed to be equivalent).

The one rule: the format and Base URL must be the same combination, and copy the address exactly (do not add or remove /v1).

4. The model ID

Select it from the ZCode model list first; if it is not there, copy a real, available ID from the 👉 DeepSeek console or the 👉 official model docs. Case and symbols must match exactly — never type it from memory.

The model name just changed — don't copy old tutorials

DeepSeek has released V4.1-Flash, and the official pricing page states to use deepseek-flash; the earlier deepseek-v4-flash and deepseek-v4-flash-vision-exp still work but are legacy names (requests route to V4.1-Flash).

What to do: use whichever ID your list shows. Do not hand-edit it just to match a tutorial.

🧩 Input Types

Edit the model: the UI offers four input switches — Text / Image / Video / PDF. Enable all four, set the output type to Text, and save.

Enabled does not mean verified

The UI offering a switch only means ZCode provides the option. Before first using a given file type, test it once with an ordinary, non-sensitive file — for example upload a random image and ask what is in it — and only use it on real material after it works. If one type fails in your test, do not rely on it.

Do not guess capability from the model name

Rely on the vendor's official model list and the capability labels shown in ZCode. Do not upload business data just to test.

If your UI shows reasoning levels off / high / max, those are ZCode UI options, not DeepSeek API parameter names; defaults and effects follow your current ZCode version, and for direct API use follow the official documentation.

✅ Test and Use It

1. Test the connection

Click Test in the model list; a green Connected means it works.

Test failed? Match the exact message first:

Insufficient Balance = your balance ran out — top up first, don't rush to change settings

This message is DeepSeek's out-of-balance error (official code 402). Top up first: add funds at the 👉 top-up page, then click Test again.

Don't keep editing the Base URL, API format, or API key just because of this message — an empty balance is not fixed by changing configuration. If it still won't connect after topping up, work through the table below for the model, permissions, or service status.

For other test failures, match here:

Message	Cause	What to do
"Model not found"	The ID is wrong, retired, or not offered to this account / API format	Copy the current available ID from the ZCode list or console; if it still fails, check whether the model supports the selected format and your account permissions
Other connection failures	Could be an invalid key, format/address mismatch, permissions, rate limits, network, or a server hiccup	Check the full error first: 401 → check the key, 429 → stagger or retry later, 5xx → retry later; if the message is unclear, recheck the format and Base URL

Then send one text-only message (for example "Explain what a sales inquiry is in one sentence"); a normal answer means the basic setup is done. If you need image capability, send an ordinary test image and ask what is in it.

2. How to select it

Below the prompt box in a new task there is a model menu; choose DeepSeek → the model you configured.

3. Add it to the team

If you installed the ZCode AI agent team, installation or update reads the capability list and assigns roles automatically. DeepSeek may be assigned to reasoning-heavy roles — follow the role list after install/update. To reassign: confirm the provider is enabled → copy the current "update" prompt from the agent-team repository → explicitly ask to reassign → start a new session and check the roles.

📋 Common Mistakes

Pitfall	Consequence	How to avoid
🔴 Leaking the API key	Someone else can spend your balance	Keep it only in ZCode; never paste it into chats/screenshots/repos; disable and recreate it if usage looks odd
🟡 Wrong or legacy model ID	"Model not found" error	Copy the current available ID from the ZCode list or console
🟡 Format and Base URL mismatch	Connection fails	Use the matching pair; copy the address exactly
🟡 Treating a UI switch as capability	Video/PDF upload fails or is unreadable	An enabled switch is not a guarantee; confirm official capability, then test each item
🟢 No input types enabled	Model cannot read uploaded images	Enable all four; test each type with an ordinary file before first use
🟡 Balance exhausted (test shows Insufficient Balance)	Requests rejected; long tasks cut off midway	Top up at the 👉 top-up page and test again; if concurrency-limited, stagger or wait

✅ Completion Check

• DeepSeek account registered and signed in;
• API key created, and confirmed it was not pasted anywhere public;
• A small top-up made (or balance confirmed sufficient);
• The DeepSeek provider in ZCode shows Enabled;
• The API format and Base URL are the same combination;
• The model ID was copied from the list or console;
• All four input types enabled, with each type tested once before first use;
• Clicking Test shows Connected, and the text test passes.

🚀 Next

📄 Step 4 · Install the ZCode AI Agent Team — put DeepSeek to work alongside other models by role.

📚 Further Reading

• 📚 Step 3 · Configure Plans and Models: the Base URL cheat sheet for other providers
• 📚 Build a B2B Inquiry Workflow with ZCode and Laifaxin: use these models for real business
• 📚 DeepSeek official pricing page: the latest billing rules