DeepSeek

Open Interpreter connects directly to the DeepSeek API with the built-in deepseek provider. It uses DeepSeek's OpenAI-compatible Chat endpoint and reads your key from DEEPSEEK_API_KEY.

Start With DeepSeek

Create a key in the DeepSeek platform, export it, and start Open Interpreter:

export DEEPSEEK_API_KEY="..."
interpreter

Open /model, select DeepSeek, then select a model. For a direct launch:

DEEPSEEK_API_KEY="..." interpreter \
  -c 'model_provider="deepseek"' \
  -m deepseek-v4-pro

For one non-interactive task:

DEEPSEEK_API_KEY="..." interpreter exec \
  -c 'model_provider="deepseek"' \
  -m deepseek-v4-pro \
  "Review this repository and fix the highest-impact bug."

Choose a Model

The /model picker is generated from maintained provider catalogs and the provider's current model data. Use it as the source of truth for models available in your installed version.

DeepSeek's current API model IDs are deepseek-v4-pro and deepseek-v4-flash. Use Pro for the higher-capacity option and Flash when cost or concurrency matters more. DeepSeek has announced that the legacy deepseek-chat and deepseek-reasoner IDs will be discontinued on July 24, 2026, so new configurations should use the V4 IDs. See DeepSeek's API updates for current availability and its pricing page before committing to a workload.

Harness Behavior

DeepSeek models use claude-code-bare automatically when you have not set a harness. Open Interpreter carries that smaller Claude Code-shaped agent surface over DeepSeek's Chat Completions endpoint; it does not run an external CLI.

To inspect or change the active harness, run /harness. The optional deepseek-tui mode is available when you specifically want its DeepSeek TUI/CodeWhale-shaped prompt and tools:

model_provider = "deepseek"
model = "deepseek-v4-pro"
harness = "deepseek-tui"

Remove the explicit harness line to return to the recommended automatic default.

Configuration

The built-in provider is equivalent to this connection:

model_provider = "deepseek"
model = "deepseek-v4-pro"

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

You normally do not need to copy that provider block. It is useful when you need to understand a proxy or custom deployment. If you change the endpoint, keep wire_api = "chat" for DeepSeek's OpenAI-compatible Chat API.

Editors and SDKs

The same provider configuration works through interpreter acp in ACP-compatible editors and through Open Interpreter's Codex SDK compatibility.

Troubleshooting

  • A 401 or 403 usually means DEEPSEEK_API_KEY is missing, expired, or belongs to a different endpoint.
  • If a legacy model ID stops working, open /model and select a current V4 model instead of hard-coding the replacement.
  • If the agent surface is unexpected, run /harness and remove any old explicit harness value from ~/.openinterpreter/config.toml.
  • If a proxy exposes a different protocol, match its wire_api; do not assume every DeepSeek-compatible endpoint has the same transport.
Give feedback