How to Use DeepSeek in VS Code With Cline, Continue and Roo Code (2026)

DeepSeek is one of the cheapest strong models you can plug into VS Code. With the right extension, it can read your code, edit files and run commands for a few cents per task.

This guide shows you how to set it up in Cline, Continue and Roo Code (plus its successor, Kilo Code). It also covers model names, real costs, running DeepSeek locally with Ollama, and fixes for common errors.

Which VS Code Extension to Pick

All three tools are free, open-source extensions where you bring your own API key. You only pay DeepSeek for the tokens you use.

Quick answer: Pick Cline if you want an agent that does the work for you. Pick Continue if you want a chat sidebar and autocomplete while you type.

For a wider look at coding agents, see my Cursor vs Windsurf vs Claude Code vs Cline comparison.

Step 1: Get a DeepSeek API Key

The DeepSeek API is prepaid. You add credit first, and each call is taken from that balance.

1. Sign in. Go to platform.deepseek.com and sign up or log in.
2. Top up. Open the Top up page and add a small amount, such as $5. With a zero balance, every request fails with error 402.
3. Create a key. Open API Keys, click Create new API key, and name it “vscode” or similar.
4. Copy it right away. DeepSeek shows the key only once. Store it in a password manager.

Warning: Never paste your key into a file that goes into git. If a key leaks, delete it on the API Keys page and create a new one.

If you also want to call DeepSeek from your own scripts, my DeepSeek API Python tutorial walks through the same key setup in code.

Step 2: Choose the Right DeepSeek Model

DeepSeek’s model names changed several times in 2026, and many older guides still show the retired names. Here is the current list from DeepSeek’s API docs.

Key 2026 changes in one list

• September 10, 2026: V4.1 Flash launched as deepseek-flash with image understanding and lower prices.
• October 17, 2026: deepseek-flash was upgraded to V4.1-Flash-1017 with no change to calls or prices.
• Thinking mode: it is on by default. The model reasons before it answers, which helps with code but adds output tokens.

A simple rule: start every project on deepseek-flash. Switch to deepseek-v4-pro only for a task where Flash keeps getting it wrong, then switch back.

How to Set Up DeepSeek in Cline

Cline is an agent. It plans a task, edits files and runs terminal commands, asking for your approval along the way.

1. Install Cline. In VS Code, open Extensions (Ctrl+Shift+X, or Cmd+Shift+X on Mac), search for “Cline” and click Install.
2. Open settings. Click the Cline icon in the sidebar, then the gear icon at the top of the panel.
3. Pick the provider. In the API Provider dropdown, choose DeepSeek.
4. Paste your key. Put it in the DeepSeek API Key field.
5. Pick a model. In the Model dropdown, choose deepseek-flash. Choose deepseek-v4-pro for heavier work.
6. Test it. Open a project and ask: “Explain what this repo does and list its main files.” Cline should read files and reply.

If the model you want is missing

Extension model lists sometimes lag behind DeepSeek’s changes. Use Cline’s OpenAI Compatible provider instead and fill in three fields:

• Base URL: https://api.deepseek.com
• API Key: your DeepSeek key
• Model ID: deepseek-flash or deepseek-v4-pro

Note: A public Cline issue (#14538) reported that the older deepseek-v4-flash entry showed a 128K context and $0.00 price. Pick deepseek-flash so the context and cost display are right.

How to Set Up DeepSeek in Continue

Continue gives you a chat panel, inline edits (Ctrl+I or Cmd+I) and tab autocomplete. Models are set in a config.yaml file.

1. Install Continue. Search for “Continue” in the Extensions view and install it.
2. Open the config. Click the Continue icon, open the model selector or settings, and choose to edit config.yaml. Locally it lives at ~/.continue/config.yaml.
3. Add DeepSeek models. Paste this under models, using your own key:

   name: My Config
   version: 0.0.1
   schema: v1
   
   models:
     - name: DeepSeek Flash
       provider: deepseek
       model: deepseek-flash
       apiKey: YOUR_DEEPSEEK_API_KEY
       roles:
         - chat
         - edit
         - apply
     - name: DeepSeek V4 Pro
       provider: deepseek
       model: deepseek-v4-pro
       apiKey: YOUR_DEEPSEEK_API_KEY
       roles:
         - chat

4. Save and pick the model. Save the file. Choose “DeepSeek Flash” from the model dropdown in the chat panel.
5. Test it. Highlight a function, press Ctrl+L (Cmd+L on Mac) to send it to chat, and ask for a review.

What about autocomplete?

Continue’s DeepSeek page still lists the older deepseek-coder name for autocomplete. Autocomplete fires on almost every keystroke, so a big remote model is slow and wasteful for it.

A small local model through Ollama is usually the better fit. See the Ollama section below for a ready config.

Roo Code and Kilo Code

Roo Code announced in April 2026 that it would close its VS Code extension. Its repo was archived and the last extension release came out on May 15, 2026.

If you still have Roo Code installed

Bring-your-own-key setups keep working, but the extension gets no more updates. Its DeepSeek dropdown only knows the old names, so use the OpenAI Compatible route:

1. Open Roo settings. Click the gear icon in the Roo Code panel.
2. Choose the provider. Set API Provider to OpenAI Compatible.
3. Fill the fields. Base URL https://api.deepseek.com, your API key, and model ID deepseek-flash.
4. Save and test. Ask it to summarize one file to confirm the connection.

Moving to Kilo Code

Kilo Code started as a fork of Roo and lists DeepSeek as a provider. Its migration guide says API keys do not transfer, so you add your DeepSeek key again under the gear icon, then Providers.

• Rules: rename .roorules to AGENTS.md and copy .roo/rules to .kilo/rules.
• Modes: copy .roomodes to .kilocodemodes.
• MCP: copy .roo/mcp.json to .kilocode/mcp.json.

What DeepSeek Costs in VS Code

DeepSeek charges per million tokens. Since August 2026 it uses peak and off-peak prices, and off-peak is half price. These rates were last confirmed unchanged by price trackers on October 29, 2026.

• Peak hours: 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday.
• Off-peak: all other hours, plus weekends and Chinese public holidays.
• Caching: automatic. Agents resend the same files and system prompt, so many input tokens become cheap cache hits.

A worked example

Say a Cline task sends 1M input tokens, 80% of them cache hits, and gets back 50,000 output tokens on deepseek-flash at peak. That is about $0.005 + $0.06 + $0.06, so roughly $0.13.

The same task on deepseek-v4-pro would be about $0.035 + $0.26 + $0.20, or around $0.50. For the full breakdown, read my DeepSeek API pricing guide.

Run DeepSeek Locally With Ollama

If you want your code to stay on your machine, run a DeepSeek model with Ollama. The trade-off is quality: local models are much smaller than the API models.

Note: The small deepseek-r1 sizes are distilled models built on Qwen and Llama, not the full DeepSeek model. The :cloud tag runs on Ollama’s servers, so your code does leave your machine.

Set it up

1. Install Ollama and pull a model: ollama pull deepseek-r1:14b
2. Raise the context size. Agents send a lot of text. Set a bigger context in the Ollama app settings, or start the server with OLLAMA_CONTEXT_LENGTH=32768 ollama serve.
3. Cline: set API Provider to Ollama, keep the base URL as http://localhost:11434, and choose the model.
4. Continue: add local models to config.yaml:

     - name: DeepSeek R1 14B (local)
       provider: ollama
       model: deepseek-r1:14b
       roles:
         - chat
         - edit
     - name: DeepSeek Coder autocomplete
       provider: ollama
       model: deepseek-coder:1.3b
       roles:
         - autocomplete

For other local picks by GPU size, see my best Ollama models guide.

Troubleshooting DeepSeek in VS Code

The 400 error needs a short explanation. With tools enabled, DeepSeek requires the earlier reasoning_content to be sent back on every turn, and some extension versions do not do this yet.

If you see lots of 503s, the cheapest fix is often timing. Running heavy agent jobs outside the peak windows also halves your bill.

Frequently Asked Questions

Can I use DeepSeek for free in VS Code?

The extensions are free, but the DeepSeek API is prepaid. For zero cost, run a local model like deepseek-r1:8b through Ollama.

Which DeepSeek model is best for coding?

Use deepseek-flash for everyday coding because it is fast and cheap. Use deepseek-v4-pro for the hardest tasks.

Is deepseek-chat still working?

No. deepseek-chat and deepseek-reasoner were scheduled for retirement on July 24, 2026 and no longer appear in DeepSeek’s docs. Use deepseek-flash.

Is Cline or Continue better with DeepSeek?

Cline is better for hands-off agent work like multi-file changes. Continue is better for chat, quick edits and autocomplete while you type.

Does Roo Code still work?

The extension was archived on May 15, 2026, and gets no updates. It can still connect with your own key through OpenAI Compatible mode, but Kilo Code is the maintained alternative.

What is the DeepSeek base URL for VS Code extensions?

Use https://api.deepseek.com for OpenAI-compatible settings. DeepSeek also offers https://api.deepseek.com/anthropic for tools that use the Anthropic format.

Is my code sent to DeepSeek?

Yes, when you use the API, the files your agent reads are sent to DeepSeek’s servers. Use a local Ollama model if your code must stay on your machine. My DeepSeek vs Claude guide compares the options.

Sources

• DeepSeek API Docs: Your first API call and model names
• DeepSeek API Docs: Change log
• Cline Docs: DeepSeek provider
• Continue Docs: DeepSeek provider
• Roo Code Docs: Using DeepSeek
• The New Stack: Roo Code shuts down its VS Code extension
• Kilo Code: Roo to Kilo migration guide
• BenchLM: DeepSeek API pricing tracker
• Developers Digest: DeepSeek in Cline setup notes
• Ollama Library: deepseek-r1
• Ollama Library: deepseek-v4.1-flash