Use the Wokey API with DeepSeek Harness

1. 1

   Install the 0.2.1 pre-release

   npm’s latest tag is still 0.2.0; DeepSeek recommends v0.2.1-rc.1 or later for V4.1-Flash-1017, so install @next.

2. 2

   Set the API key variable

   The config only names apiKeyEnv, so the key never lands in the file. You can also save the key under Settings → Models in the web UI.

3. 3

   Add the provider

   Edit ~/.dsh/profiles/web/cordis.patch.yml (profiles/headless for dsh headless). This example uses Anthropic Messages for both models, with https://api.wokey.ai as the base URL. If you choose OpenAI Chat Completions instead, create a separate provider and use https://api.wokey.ai/v1.

4. 4

   Declare image input

   DeepSeek Harness treats an undeclared custom model as text-only, so it can reject an image before Wokey sees the request even when the Wokey model supports vision. For Opus 5.5, Haiku 5.5, or any other vision model, add input: [text, image] under that model; in the UI use Settings → Models → edit the provider → Customized settings → expand the model row → Input types → Image. Save, then start a new session.

5. 5

   Keep existing providers

   This override replaces the whole llm-pi-ai entry config, so if you already added providers in the UI, list them under providers too. Use spaces for every input indent and do not mix tabs into the YAML. Start a new session after saving.

# 1. Install or update to 0.2.1
npm install -g @deepseek-ai/dsh@next
export WOKEY_API_KEY="YOUR_API_KEY"

# 2. ~/.dsh/profiles/web/cordis.patch.yml
- id: llm-pi-ai
  config:
    providers:
      wokey:
        displayName: Wokey
        apiKeyEnv: WOKEY_API_KEY
        api: anthropic-messages
        baseURL: https://api.wokey.ai
        models:
          - id: claude-opus-5-5
            name: Claude Opus 5.5
            contextWindow: 1000000
            input: [text, image]
          - id: deepseek-v4.1-flash
            name: DeepSeek-V4.1-Flash
            contextWindow: 1000000
            maxTokens: 393216
            input: [text, image]
- id: agent-default-model
  config:
    provider: wokey
    model: claude-opus-5-5

# 3. Start
dsh web

Where it is configured: ~/.dsh/profiles/web/cordis.patch.yml

Wokey provides the model inference interfaces listed on this page, not standalone /v1/tools/search or /v1/tools/fetch endpoints. Tool calling support does not imply support for every upstream built-in web tool. Server-side search within model requests depends on the actual supply route and is not currently listed as a verified, uniformly supported Wokey capability; this page publishes no Wokey per-call Search/Fetch price. For documented standalone search and fetch support, use a DeepSeek API key with api.deepseek.com, configured and billed separately from Wokey model calls.

• “The current model does not support images”: this is Harness’s local capability gate; it does not mean Wokey or Opus 5.5 lacks vision. Enable Image for that model, or add input: [text, image] to its YAML entry.
• Images are still rejected after saving: start a new session and select the model again; an existing session keeps its recorded model capability and attachments.
• The image request returns 400: match the protocol and base URL. Anthropic Messages uses https://api.wokey.ai without /v1; OpenAI Chat Completions uses https://api.wokey.ai/v1. deepseek-v4.1-flash supports both formats, but the client still has to send the shape for the selected protocol.
• The app will not start or reports a YAML line number: check that input is nested under the intended model, every model dash is aligned, and the file contains no tabs. Restore a backup, then save it again with spaces only.
• The app starts with “Select a model”: check that agent-default-model still points to a provider and model that exist. Update it after deleting or renaming a provider.
• The model appears after discovery but has no image capability: Fetch available models mainly fills IDs and capacity fields. Explicitly declare input: [text, image] for each custom vision model instead of marking every model as vision-capable.
• Other providers disappeared after saving: a cordis.patch.yml override replaces the whole llm-pi-ai entry, so keep all existing providers under providers.

DeepSeek Harness uses the OpenAI Chat Completions format by default. A 200 from this request means the key and base URL are right. Replace YOUR_API_KEY with your key before sending it.

curl https://api.wokey.ai/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4.1-flash",
    "messages": [
      { "role": "user", "content": "Hello" }
    ]
  }'

Should the DeepSeek Harness baseURL end in /v1?

It depends on the provider's api field. With anthropic-messages use https://api.wokey.ai with no /v1; OpenAI Chat Completions needs a separate provider with https://api.wokey.ai/v1.

Why does it say the current model does not support images?

That is a local capability check in the Harness: custom models declare text input only by default. Add input: [text, image] to the model entry, save, and try again in a new session.

Why did my other providers disappear after saving?

The llm-pi-ai override in cordis.patch.yml replaces the whole entry. Write the providers you already had into providers alongside Wokey.