DeepSeek

Overview

DeepIntShield connects to DeepSeek’s OpenAI-compatible API with the provider prefix deepseek/. The adapter uses bearer authentication, performs model discovery only when /v1/models is called, and sends inference directly to the documented upstream operation without a discovery lookup on the inference path.

Supported operations

Operation	Non-streaming	Streaming	DeepIntShield route
List models	✅	—	GET /v1/models
Text completions	✅	✅ SSE	POST /v1/completions
Chat Completions	✅	✅ SSE	POST /v1/chat/completions
Responses	✅	✅ SSE	POST /v1/responses

Other operations are rejected locally as unsupported instead of being translated to a different API. DeepSeek does not use the Responses WebSocket mode in this adapter.

Configure DeepSeek

1. Open Workspace → AI Providers.
2. Select DeepSeek and add a DeepSeek API key.
3. Optionally restrict that key to an allow-list of model IDs.
4. Save the provider, then create or reuse a Virtual Key.

The default upstream base URL is https://api.deepseek.com. A custom network_config.base_url is available for an approved private endpoint or proxy. Keep provider secrets in the dashboard or an environment-backed secret; do not send a DeepSeek key to the public gateway request.

Chat Completions

curl https://app.deepintshield.com/v1/chat/completions \
  -H "Authorization: Bearer $DEEPINTSHIELD_VIRTUAL_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek/deepseek-v4-flash",
    "messages": [{"role": "user", "content": "Explain this design in three points."}],
    "stream": true
  }'

Chat requests support provider-documented function tools, tool choice, structured output, thinking controls, usage data, and data-only SSE. Preserve the tool-call IDs when sending tool results back on a later turn.

Responses

curl https://app.deepintshield.com/v1/responses \
  -H "Authorization: Bearer $DEEPINTSHIELD_VIRTUAL_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek/deepseek-v4-flash",
    "input": "Draft a migration checklist.",
    "stream": true
  }'

The adapter uses DeepSeek’s native /responses surface. The stored Responses lifecycle routes documented in Protocol operations are not enabled for DeepSeek. The current upstream Responses surface supports deepseek-v4-flash; deepseek-v4-pro is not supported on that operation. Use Chat Completions when selecting deepseek-v4-pro.

Model discovery and pricing

GET /v1/models?provider=deepseek returns the models available to the configured account and applies the key’s model allow-list. Model availability and public rates can change; review DeepSeek’s current model and pricing page and use DeepIntShield pricing overrides for contracted rates.

Note

DeepSeek’s current model IDs and limits are upstream-controlled. Discover the account’s live catalog rather than assuming an ID from an old example remains available.

Errors and retries

Upstream HTTP errors are normalized with the provider, requested model, request type, and safe response headers attached. Retry only errors marked retryable; do not retry invalid parameters, an unsupported operation, or a denied model.

See the official DeepSeek Chat Completions reference and model discovery reference.