How I connected 300+ AI Models to OpenClaw with Kilo Code API Key

A step-by-step guide to setting up Kilo Code API with OpenClaw — with the sync script that makes it all work.

---

A few weeks ago, I was juggling three different API keys and auths just to test various AI models. GPT-5.2, sonnet-4.5, Gemini... you get the picture. Each had its own pricing, rate limits, and authentication quirks. It was a mess and quite pricey.

Then I stumbled upon a tweet by Kilo Code mentioning GLM-5 and MiniMax 2.5 available for free. One API key, 300+ models, unified interface. Sounds perfect, right?

Well, almost. Getting it working with OpenClaw wasn't as straightforward as the docs suggested. I ran into weird context window bugs, reasoning parameter conflicts, and authentication errors that took hours to debug.

This guide is everything I wish I knew before starting. Let's save you that headache.

---

Why Kilo Code API?

Managing five different AI provider keys is annoying. Kilo solves that. One API key, 300+ models, all through a single OpenAI-compatible endpoint:

https://api.kilo.ai/api/gateway

Supports GPT, Claude, Gemini, Kimi, GLM, and hundreds more. Often cheaper than going direct. Has a generous free tier too.

---

Step 1: Get Your Kilo API Key

1. Sign up at kilo.ai
2. Go to your dashboard → API Keys
3. Generate a new key and copy it

---

Step 2: Connect Kilo to OpenClaw

Run the onboarding wizard:

openclaw onboard

When prompted, select:

• AI Provider → Custom Provider
• Base URL → https://api.kilo.ai/api/gateway
• API Type → OpenAI-compatible
• API Key → Your Kilo key from Step 1
• Model ID → e.g. anthropic/claude-sonnet-4.5
• API Alias → kilo-api

⚠️ The alias kilo-api is important — the sync script in Step 3 references this exact name in the config. Use it as-is.

At this point one model works. But to unlock all 300+, you need the next step.

---

Step 3: Sync All Models

The onboarding only registers one model. This script syncs all 300+ and fixes three bugs you would hit otherwise:

• Context window stuck at 4096 — some models have 128K–256K windows but OpenClaw defaults them down
• Reasoning parameter conflict — 7 models have mutually exclusive reasoning params, causing a 400 error
• Model not allowed — models must be in both provider list AND allowlist; onboarding only does one

Save as ~/sync-kilo-models.sh

#!/bin/bash
# sync-kilo-models.sh - Sync model metadata from Kilo API to OpenClaw config
# Models are added directly under the kilo-api provider (no prefix)

CONFIG_FILE="/home/antarikshc/.openclaw/openclaw.json"
KILO_API_KEY=$(jq -r '.models.providers."kilo-api".apiKey' "$CONFIG_FILE")

echo "Fetching models from Kilo API..."
MODELS_JSON=$(curl -s "https://api.kilo.ai/api/gateway/models" \
  -H "Authorization: Bearer $KILO_API_KEY")

# Check if we got valid data
if [ -z "$MODELS_JSON" ] || [ "$MODELS_JSON" = "null" ]; then
  echo "❌ Failed to fetch models from Kilo API"
  exit 1
fi

echo "Building OpenClaw model config..."

# Create the models array from Kilo data
# Handle reasoning parameter:
# - If model has BOTH "reasoning" AND "reasoning_effort" -> set reasoning: false (they're mutually exclusive)
# - If model has ONLY "reasoning" -> set reasoning: true
# - Otherwise -> set reasoning: false
MODELS_ARRAY=$(echo "$MODELS_JSON" | jq '[.data[] | 
  {
    "id": .id,
    "name": (.name // .id),
    "reasoning": (
      if (.supported_parameters // [] | contains(["reasoning"])) and 
         (.supported_parameters // [] | contains(["reasoning_effort"])) then
        false  # Both present - mutually exclusive, let user configure manually
      elif (.supported_parameters // [] | contains(["reasoning"])) then
        true   # Only reasoning supported
      else
        false  # No reasoning support
      end
    ),
    "input": ["text"],
    "cost": {
      "input": ((.pricing.prompt // "0") | tonumber * 1000000),
      "output": ((.pricing.completion // "0") | tonumber * 1000000),
      "cacheRead": 0,
      "cacheWrite": 0
    },
    "contextWindow": (.context_length // .top_provider.context_length // 128000),
    "maxTokens": (.top_provider.max_completion_tokens // 4096)
  }
]')

echo "Found $(echo "$MODELS_ARRAY" | jq 'length') models"

# Count reasoning models
REASONING_COUNT=$(echo "$MODELS_ARRAY" | jq '[.[] | select(.reasoning == true)] | length')
echo "  - $REASONING_COUNT models with reasoning (exclusive)"

# Count models with both reasoning + reasoning_effort (set to false)
BOTH_COUNT=$(echo "$MODELS_JSON" | jq '[.data[] | select(
  (.supported_parameters // [] | contains(["reasoning"])) and 
  (.supported_parameters // [] | contains(["reasoning_effort"]))
)] | length')
echo "  - $BOTH_COUNT models have BOTH reasoning + reasoning_effort (set to false to avoid conflict)"

# Create the agents.defaults.models entries (for allowlist)
MODELS_ALLOWLIST=$(echo "$MODELS_JSON" | jq '[
  .data[] | 
  {
    "key": ("kilo-api/" + .id),
    "value": {"alias": (.id | split("/") | last | gsub(":free"; "") | gsub(":exacto"; ""))}
  }
] | from_entries')

echo "Updating OpenClaw config..."

# Update both the provider models AND the agents.defaults.models
jq --argjson models "$MODELS_ARRAY" --argjson allowlist "$MODELS_ALLOWLIST" '
  .models.providers."kilo-api".models = $models |
  .agents.defaults.models = $allowlist
' "$CONFIG_FILE" > "${CONFIG_FILE}.tmp" && mv "${CONFIG_FILE}.tmp" "$CONFIG_FILE"

echo "✅ Config updated!"
echo ""
echo "Sample models synced:"
echo "$MODELS_ARRAY" | jq -r '.[] | "  - \(.id): contextWindow=\(.contextWindow), reasoning=\(.reasoning)"' | head -10
echo "  ... and $(echo "$MODELS_ARRAY" | jq 'length - 10') more"
echo ""
echo "Allowlist entries: $(echo "$MODELS_ALLOWLIST" | jq 'length')"

Run It

chmod +x ~/sync-kilo-models.sh
~/sync-kilo-models.sh

✅ Expected: Found 319 models ... Done! 319 models synced.

---

Step 4: Restart and Verify

openclaw gateway restart

Then test a couple of models:

/model kilo-api/moonshotai/kimi-k2.5
Hey, working?

/model kilo-api/anthropic/claude-opus-4.6
How about you?

If both respond, you're done! 🎉

You will see all the Kilo models with a kilo-api/ prefix.

---

Free Models Worth Knowing

• z-ai/glm-5:free — 202K context, general + reasoning
• minimax/minimax-m2.5:free — 204K context, coding & productivity
• stepfun/step-3.5-flash:free — 256K context, fast responses
• arcee-ai/trinity-large-preview:free — 131K context, creative writing

---

Quick Reference

# Switch model
/model kilo-api/moonshotai/kimi-k2.5

# Check current model
/status

# Re-sync models (run monthly or after Kilo adds new ones)
~/sync-kilo-models.sh && openclaw gateway restart