Set up custom AI model API

[mgitlabs]

I have an LLM model and want to add it to OpenClaw, but I don't know how to configure it because I can't access the OpenClaw CLI. Please guide me.

curl -X POST https://**********.workers.dev \
  -H "Authorization: Bearer ***********" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Write a short poem about a rainy day.",
    "systemPrompt": "You are a helpful assistant.",
    "history": [
      { "role": "user", "content": "Tell me something about weather." },
      { "role": "assistant", "content": "Weather is the state of the atmosphere." }
    ]
  }'

[ay0400]

I have an LLM model and want to add it to OpenClaw, but I don't know how to configure it because I can't access the OpenClaw CLI. Please guide me.

curl -X POST https://**********.workers.dev
-H "Authorization: Bearer ***********"
-H "Content-Type: application/json"
-d '{
"prompt": "Write a short poem about a rainy day.",
"systemPrompt": "You are a helpful assistant.",
"history": [
{ "role": "user", "content": "Tell me something about weather." },
{ "role": "assistant", "content": "Weather is the state of the atmosphere." }
]
}'

I’ll structure it like this:

1. Big Picture (how OpenClaw works)

2. General Step-by-Step (works for ANY custom model)

3. GLM 5.1 From Modal As A Case Study (your exact setup)

---

🧠 1. Big Picture — How OpenClaw Actually Works

From OpenClaw, the key idea is:

OpenClaw does NOT call models directly.
It routes through a model registry system.

Internally, it works like this:

User selects model → agent config → models.providers → API call

So for ANY model to work, it must exist in:

models.providers

---

🔑 Key Rule (Everything revolves around this)

Every model must follow this format:

provider/model-id

Example:

modal/zai-org/GLM-5.1-FP8

Where:

• modal = provider name

• zai-org/GLM-5.1-FP8 = actual model ID

---

❗ Why Custom Models Fail by Default

Because:

• OpenClaw validates models against its internal catalog

• Your custom API is NOT in that catalog

👉 So you must manually register it

---

⚙️ 2. General Setup (Works for ANY Custom AI Model)

This is the reusable system you built.

---

✅ Step 1 — Head over to HuggingClaw in your Hugging Face

• Go to Hugging Face

• Under your own version

• Open Files → start.sh

---

✅ Step 2 — Inject Custom Provider Logic

Find this line:

CONFIG_JSON=$(echo "$CONFIG_JSON" | jq ".agents.defaults.model = \"$LLM_MODEL\"")

Add this right below it:

if [ -n "$CUSTOM_PROVIDER_NAME" ] && [ -n "$CUSTOM_BASE_URL" ] && [ -n "$CUSTOM_MODEL_ID" ]; then
  echo "🔧 Registering custom provider: $CUSTOM_PROVIDER_NAME → $CUSTOM_BASE_URL"
  CONFIG_JSON=$(echo "$CONFIG_JSON" | jq \
    ".models.mode = \"merge\" |
     .models.providers[\"$CUSTOM_PROVIDER_NAME\"] = {
       \"baseUrl\": \"$CUSTOM_BASE_URL\",
       \"apiKey\": \"$LLM_API_KEY\",
       \"api\": \"openai-completions\",
       \"models\": [{
         \"id\": \"$CUSTOM_MODEL_ID\",
         \"name\": \"$CUSTOM_MODEL_ID\",
         \"contextWindow\": 128000,
         \"maxTokens\": 500
       }]
     }")
fi

---

✅ Step 3 — Create Secrets in Hugging Face

Go to:

Settings → Secrets

Create these:

Secret Name : Value

LLM_API_KEY: Your model API key
LLM_MODEL: provider/model-id
CUSTOM_PROVIDER_NAME: provider name
CUSTOM_BASE_URL: API base URL
CUSTOM_MODEL_ID: raw model ID
GATEWAY_TOKEN: internal OpenClaw auth

---

🔹 Resulting Internal Config

OpenClaw builds this:

{
  "models": {
    "mode": "merge",
    "providers": {
      "modal": {
        "baseUrl": "https://api.us-west-2.modal.direct/v1",
        "apiKey": "YOUR_KEY",
        "api": "openai-completions",
        "models": [
          {
            "id": "zai-org/GLM-5.1-FP8",
            "name": "zai-org/GLM-5.1-FP8"
          }
        ]
      }
    }
  },
  "agents": {
    "defaults": {
      "model": "modal/zai-org/GLM-5.1-FP8"
    }
  }
}

---

🔹 What Happens When You Send a Prompt

1. You select:

   modal/zai-org/GLM-5.1-FP8
   

2. OpenClaw:

   • finds provider modal

   • finds model zai-org/GLM-5.1-FP8

3. Sends request to:

   https://api.us-west-2.modal.direct/v1/chat/completions
   

4. Includes:

   Authorization: Bearer YOUR_KEY
   

👉 Exactly like your curl command

---

⚠️ Common Errors (And Fixes)

❌ “Invalid api type”

→ Use:

openai-completions

---

❌ “Model not found”

→ mismatch between:

• LLM_MODEL

• CUSTOM_MODEL_ID

---

❌ Silent failure

→ missing:

OPENAI_API_KEY mapping

---

❌ Doesn’t show in UI

→ model not registered in models.providers

---