DeepSeek Harness

DeepSeek Harness (dsh) is the recommended coding harness for the model Inferam serves today. It talks to Inferam over its DeepSeek route, which uses the Anthropic Messages format.

With the CLI

inferam dsh            # the web profile: dsh's browser UI
inferam dsh tui        # another profile
inferam dsh headless "run the tests"

dsh needs Node.js 22 or later. If it isn’t installed, the CLI offers npm install -g @deepseek-ai/dsh, its only official install channel.

For each run, the CLI passes dsh a --patch overlay file. It ranks above your profile and home patches, and is deleted when dsh exits. The overlay:

• points dsh’s DeepSeek route at https://api.inferam.ai with your Inferam key and the served model;
• turns off session-log upload (session-log-deepseek), which would otherwise send session content to DeepSeek;
• sets DSH_TELEMETRY_DISABLED=1 to turn off feedback telemetry.

With the CLI, your sessions stay between you and Inferam.

By hand

Save this as inferam-patch.json:

[
  {
    "id": "llm-deepseek",
    "config": {
      "baseURL": "https://api.inferam.ai",
      "apiKeyEnv": "INFERAM_API_KEY",
      "models": [
        { "id": "DeepSeek-V4.1-Flash-1017", "name": "DeepSeek-V4.1-Flash-1017", "contextWindow": 524288, "maxTokens": 32768, "inputModalities": ["text"] }
      ]
    }
  },
  { "id": "agent-default-model", "config": { "provider": "deepseek-official", "model": "DeepSeek-V4.1-Flash-1017" } },
  { "id": "session-log-deepseek", "config": { "enabled": false } }
]

Then:

export INFERAM_API_KEY=ifr_...
export DSH_TELEMETRY_DISABLED=1
dsh web --patch ./inferam-patch.json

--patch goes right after the profile name.