DeepSeek V4 Agent Setup: OpenCode, Codex, Copilot CLI, Cline, Kilo

DeepSeek V4 is easiest to use in agents that support either OpenAI-compatible chat completions or Anthropic-compatible messages. The two base URLs are:

OpenAI compatible:    https://api.deepseek.com
Anthropic compatible: https://api.deepseek.com/anthropic

Use deepseek-v4-flash for cheap high-volume work and deepseek-v4-pro for harder coding or long-horizon agent tasks. DeepSeek’s current coding-agent guide recommends V4 Pro for OpenCode reasoning-heavy work.

Official references: DeepSeek agent integrations, DeepSeek Anthropic API, and DeepSeek pricing.

Shared Values

Setting	Value
API key	Your DeepSeek API key from `platform.deepseek.com`
OpenAI base URL	`https://api.deepseek.com`
Anthropic base URL	`https://api.deepseek.com/anthropic`
Cheap model	`deepseek-v4-flash`
Stronger model	`deepseek-v4-pro`

OpenCode

Create or update ~/.config/opencode/opencode.jsonc:

{
  "provider": {
    "deepseek": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "DeepSeek",
      "options": {
        "baseURL": "https://api.deepseek.com",
        "apiKey": "{env:DEEPSEEK_API_KEY}"
      },
      "models": {
        "deepseek-v4-pro": {
          "name": "DeepSeek-V4-Pro",
          "limit": {
            "context": 1048576,
            "output": 262144
          },
          "reasoning": true,
          "options": {
            "reasoningEffort": "max",
            "thinking": {
              "type": "enabled"
            }
          }
        }
      }
    }
  }
}

Then run:

export DEEPSEEK_API_KEY=your-deepseek-api-key
opencode

Use deepseek-v4-flash as an additional custom model only when cost matters more than coding quality.

Codex CLI

Use this only if your Codex CLI version still supports Chat Completions providers. If your version requires the Responses API only, prefer OpenCode or Claude Code for DeepSeek V4.

~/.codex/config.toml:

model = "deepseek-v4-flash"
model_provider = "deepseek"

[model_providers.deepseek]
name = "DeepSeek"
env_key = "DEEPSEEK_API_KEY"
base_url = "https://api.deepseek.com"
wire_api = "chat"

Then set the key:

export DEEPSEEK_API_KEY=your-deepseek-api-key
codex

Cline

In the Cline extension or CLI, choose an OpenAI-compatible provider:

Field	Value
API Provider	OpenAI Compatible
Base URL	`https://api.deepseek.com`
API Key	Your DeepSeek API key
Model ID	`deepseek-v4-flash` or `deepseek-v4-pro`
Context Window	`1048576`

For Cline CLI, a typical command is:

cline auth -p openai -k your-deepseek-api-key -b https://api.deepseek.com -m deepseek-v4-flash

Kilo Code and Roo Code

Both tools usually work through an OpenAI-compatible custom provider:

Provider: OpenAI Compatible
Base URL: https://api.deepseek.com
API Key: your-deepseek-api-key
Model: deepseek-v4-flash

If the tool has a separate “supports images” toggle, turn it off for DeepSeek V4. If it has a context window field, use 1048576.

GitHub Copilot CLI

DeepSeek’s official Copilot CLI guide recommends the Anthropic provider mode, not the OpenAI provider mode, because thinking-mode requests need Anthropic-compatible message handling.

export COPILOT_PROVIDER_TYPE=anthropic
export COPILOT_PROVIDER_BASE_URL=https://api.deepseek.com/anthropic
export COPILOT_PROVIDER_API_KEY=your-deepseek-api-key
export COPILOT_MODEL=deepseek-v4-pro
copilot

If your Copilot CLI does not know the DeepSeek V4 token limits, set them explicitly:

export COPILOT_PROVIDER_MAX_PROMPT_TOKENS=840000
export COPILOT_PROVIDER_MAX_OUTPUT_TOKENS=128000

Deep Code

DeepSeek also documents Deep Code, an open-source terminal coding assistant built around DeepSeek V4.

{
  "env": {
    "MODEL": "deepseek-v4-pro",
    "BASE_URL": "https://api.deepseek.com",
    "API_KEY": "sk-..."
  },
  "thinkingEnabled": true,
  "reasoningEffort": "max"
}

Then run deepcode inside your project.

OpenClaw

For OpenClaw, re-run onboarding and select DeepSeek as the model provider:

openclaw onboard --install-daemon

When prompted, enter your DeepSeek API key and use deepseek-v4-pro or deepseek-v4-flash as the default model.

Cost Guardrails

DeepSeek cache hits are the reason agent bills can be surprisingly low. Keep repository rules, system prompts, and stable context at the beginning of the prompt so the prefix can be reused. Avoid injecting timestamps, random IDs, or constantly changing logs before the stable prefix, because that can reduce cache hits.