DeepSeek Harness

DeepSeek Harness (dsh) is DeepSeek AI’s open-source agent harness. Register ApiGo as a custom llm-pi-ai provider so Harness uses your ApiGo key, model catalog, usage controls, and call logs.

DeepSeek Harness is currently in Developer Preview and may introduce compatibility-breaking changes.

Prerequisites

Node.js installed
An ApiGo API key from Workspace → API Keys
A current model slug from the Model Catalog

1. Store the ApiGo key

Create ~/.dsh/.credentials.yaml:

APIGO_API_KEY: sk-apg-...

Restrict access to the Harness directory and credential file:

chmod 700 ~/.dsh
chmod 600 ~/.dsh/.credentials.yaml

Keep the key out of settings.yaml, project files, shell history, and Git. APIGO_API_KEY is the credential name referenced by the provider configuration below.

2. Register ApiGo as a provider

Create or edit ~/.dsh/settings.yaml:

ui-onboarding:
  welcomeNoticeVersion: 2026-08-13.1
agent-presets:
  default: code
llm-pi-ai:
  providers:
    apigo:
      displayName: apigo
      apiKeyEnv: APIGO_API_KEY
      api: openai-completions
      baseURL: https://api.apigo.ai/v1
      models:
        - id: deepseek-v4-flash
          name: deepseek-v4-flash
        - id: deepseek-v4-pro
          name: deepseek-v4-pro
agent-default-model:
  provider: apigo
  model: deepseek-v4-flash

This configuration is required because ApiGo is a custom provider route. Setting DEEPSEEK_BASE_URL alone leaves Harness on its native deepseek-official route and does not select the ApiGo provider.

Use only model slugs currently shown in the ApiGo catalog. Change both the provider’s models list and agent-default-model.model when selecting another model.

3. Start Harness

Run the Web UI:

npx @deepseek-ai/dsh web

Open http://127.0.0.1:3080, create a session, and confirm that the selected provider and model are apigo / deepseek-v4-flash.

For a one-shot terminal check:

npx @deepseek-ai/dsh --profile headless \
  'Reply with exactly APIGO_HARNESS_OK and nothing else.'

A successful setup prints:

APIGO_HARNESS_OK

Then confirm the request in Workspace → Call Logs.

Configuration reference

Setting	Value	Purpose
Credential file	`~/.dsh/.credentials.yaml`	Stores the ApiGo key outside general settings
Credential name	`APIGO_API_KEY`	Matches `apiKeyEnv` in the provider profile
Provider route	`apigo`	Harness provider ID selected by `agent-default-model`
API protocol	`openai-completions`	Uses ApiGo’s OpenAI-compatible Chat Completions API
Base URL	`https://api.apigo.ai/v1`	ApiGo OpenAI-compatible gateway prefix
Default model	`deepseek-v4-flash`	Exact ApiGo model catalog slug

Troubleshooting

Symptom	Fix
`MISSING_CREDENTIAL`	Confirm `~/.dsh/.credentials.yaml` contains `APIGO_API_KEY` and is valid YAML
401 / 403	Replace the key with an active ApiGo key and check its workspace permissions or spend cap
`UNKNOWN_MODEL`	Add the exact ApiGo slug under `llm-pi-ai.providers.apigo.models`
Requests still use `deepseek-official`	Set `agent-default-model.provider` to `apigo` and restart Harness
`workspace_tpm_exceeded`	Agent prompts and tool loops can use multiple large requests; raise the Workspace TPM limit or wait for capacity before retrying
No request in Call Logs	Confirm the Base URL includes `/v1` and the active model shows provider `apigo`

Resources

Cursor

Hermes Agent