Introducing Pilot API: Smart AI Model Selection for Agents

TL;DR: Pilot API is a heuristic endpoint that helps AI agents pick the right model automatically. No more reading 1000+ lines of docs. One request, optimal model, done.

The Problem We're Solving

AI agents waste enormous context windows just figuring out which model to use.

Consider what happens today:

• OpenClaw users load 1,300+ lines of reference.md every session
• Developers spend $400+ testing different models to find the right one
• Agents burn tokens on boilerplate before doing any real work

This is expensive. This is slow. There has to be a better way.

Introducing Pilot API

Pilot API is a heuristic endpoint — it guides agents step-by-step instead of dumping documentation on them.

How It Works

Step 1: Empty request → Usage guide

curl -X POST https://api.skillboss.co/v1/pilot \
  -H "Authorization: Bearer $SKILLBOSS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{}'

Returns:

• Available model types (chat, image, video, TTS, STT, music...)
• Usage examples for each mode
• Next steps guidance

Step 2: Specify type → Smart recommendations

curl -X POST https://api.skillboss.co/v1/pilot \
  -H "Authorization: Bearer $SKILLBOSS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"type": "image", "prefer": "price", "limit": 3}'

Returns:

• Top 3 cheapest image models
• Price tier (cheap/medium/expensive)
• Complete curl examples
• Request/response documentation
• Tips and warnings for each model

Step 3: Add inputs → One-shot execution

curl -X POST https://api.skillboss.co/v1/pilot \
  -H "Authorization: Bearer $SKILLBOSS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "type": "image",
    "prefer": "price",
    "inputs": {"prompt": "A sunset over mountains"}
  }'

Auto-selects the optimal model AND executes. One request. Done.

Key Features

Smart Model Routing

Tell Pilot what you prioritize:

• "prefer": "price" — Cheapest option first
• "prefer": "quality" — Best performance first
• "prefer": "balanced" — Equal weight (default)

Capability-Based Filtering

Need something specific? Just describe it:

{
  "type": "image",
  "capability": "style transfer from reference images",
  "prefer": "quality"
}

Pilot uses semantic matching to find models that fit your needs.

Multi-Step Chains

Plan entire workflows in one request:

{
  "chain": [
    {"type": "stt", "prefer": "price"},
    {"type": "chat", "capability": "translate to Chinese"},
    {"type": "tts", "prefer": "quality"}
  ]
}

Returns recommended model for each step, data piping hints, and total cost estimate.

Intelligent Routing

Pilot automatically routes to the best option based on your preferences:

Available Models

Pilot API gives you access to 679+ endpoints across categories:

Why "Heuristic" API?

Traditional APIs are designed for human developers:

• Read docs → Pick model → Write code

Heuristic APIs are designed for AI agents:

• Send intent → Get recommendations → Execute

Every Pilot response includes guidance.next_steps — the API literally tells the agent what to do next. This is AI helping AI.

Pricing

Pilot API is included with your SkillBoss subscription. No extra cost.

• API calls: Free (Pilot is a routing layer)
• Model usage: Pay-as-you-go for actual model calls
• No rate limits: Use as much as you need

Get Started

1. Get API key at skillboss.co/console — $2 to start

For AI agents: set up skillboss.co/skill.md auto-configures SkillBoss with 687+ APIs.

2. Copy your API key from the dashboard
3. Try Pilot:

curl -X POST https://api.skillboss.co/v1/pilot \
  -H "Authorization: Bearer $SKILLBOSS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{}'

FAQ

Does Pilot replace /v1/chat/completions?

No. Pilot is for model discovery and selection. Once you know which model to use, you can call it directly via standard endpoints.

Can I still specify exact models?

Yes. Pilot is optional. You can always call models directly by ID.

What if Pilot picks a model I don't want?

Use the exclude parameter or specify model_id directly.

Is Pilot available for all SkillBoss users?

Yes. It's live now at POST /v1/pilot.

What's Next

We're working on:

• Execute mode: Auto-execute with the selected model
• Budget constraints: "Stay under $5 for this workflow"
• Agent memory: Learn from past model choices
• Health monitoring: Real-time model availability

Related Articles

• OpenClaw Alternatives: Avoid $600 Monthly Bills
• Unified AI API Gateway 2026