Nests

A Nest is a named selection of LLMs that becomes a virtual model. You pick the models, ANT routes intelligently within that selection.

Key Concept

The system prompt, JSON schema, and any domain logic live in your app. ANT passes your system prompt through unchanged to the selected model. ANT has no knowledge of what your app does — it only decides which model handles the request.

What is a Nest?

Think of a Nest as your own custom virtual model. Instead of using a preset like ant:auto that routes across all available models, a Nest restricts ANT to only the models you have hand-picked.

Example

Create a Nest called fast-coding with GPT-4o, Claude Sonnet, and Gemini Flash. Then reference it in your API requests:

model: "nest:fast-coding"

ANT scores only those three models using your Nest's preference weights and routes to the best one for each request. Every other model in the system is excluded.

Creating a Nest

Go to Dashboard → Models → Create Nest
Give your Nest a name (lowercase, hyphens allowed — e.g. fast-coding)
Select models from the catalog. Use the filters (vision, thinking, provider) to narrow down, then click Select All to quickly add every matching model
Set preference weights for Speed, Quality, and Cost to control how ANT ranks models within the Nest
Save the Nest. It is immediately available as nest:your-nest-name

Select All Shortcut

When creating a Nest, use the filter bar to narrow models by capability (vision, thinking), provider, or cost tier. Then click Select All to add every model that matches your filters in one click. This is the fastest way to build provider-specific or capability-specific Nests.

Using a Nest

Reference your Nest by passing nest:your-nest-name as the model parameter. This works with both the OpenAI and Anthropic API formats.

Python (OpenAI SDK)

from openai import OpenAI

client = OpenAI(
    base_url="https://antbase.ai/v1",
    api_key="ant_your-api-key",
)

# Route within your "fast-coding" nest
response = client.chat.completions.create(
    model="nest:fast-coding",
    messages=[
        {"role": "system", "content": "You are a senior Python developer."},
        {"role": "user", "content": "Write a Redis cache wrapper with TTL support"},
    ],
)

TypeScript

import OpenAI from 'openai';

const client = new OpenAI({
  baseURL: 'https://antbase.ai/v1',
  apiKey: 'ant_your-api-key',
});

const response = await client.chat.completions.create({
  model: 'nest:fast-coding',
  messages: [
    { role: 'system', content: 'You are a senior TypeScript developer.' },
    { role: 'user', content: 'Implement a rate limiter using the token bucket algorithm' },
  ],
});

cURL

curl https://antbase.ai/v1/chat/completions \
  -H "Authorization: Bearer ant_your-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nest:fast-coding",
    "messages": [
      {"role": "system", "content": "You are a senior developer."},
      {"role": "user", "content": "Explain the CAP theorem"}
    ]
  }'

Nest-Locked API Keys

You can create API keys that are locked to specific Nests. When a key is nest-locked, every request made with that key routes exclusively through the Nest's models — regardless of what model name the client sends.

Why Nest-Lock a Key?

Production Safety

Lock your production key to a vetted Nest so deployed apps can never accidentally route to an untested or expensive model.

Team Isolation

Give each team a key locked to their Nest. The frontend team gets fast, cheap models; the ML team gets reasoning-capable models.

Cost Control

Lock keys to Nests containing only budget-friendly models. Developers can request any virtual model and costs stay predictable.

Compliance

Restrict a key to models hosted in approved regions or by approved providers. The Nest enforces the policy automatically.

How It Works

# With a nest-locked key, the client can use any model name.
# ANT ignores it and routes within the locked nest.

response = client.chat.completions.create(
    model="ant:auto",          # ANT uses the nest's preferences instead
    messages=[{"role": "user", "content": "Hello!"}],
)

# Requesting a specific model works only if it's in the nest.
# If it's not in the nest, ANT returns an error.
response = client.chat.completions.create(
    model="gpt-4o",            # Must be in the locked nest
    messages=[{"role": "user", "content": "Hello!"}],
)

Projects

Nests live inside Projects. A Project groups related Nests and API keys together, giving you a clean organizational boundary.

One Project per app or environment. For example, create a “Production” project with a cost-controlled Nest and a “Development” project with a Nest that includes experimental models.

Each Project has its own set of API keys. Keys in a Project can be locked to any Nest within that Project. This keeps credentials, routing rules, and usage telemetry organized by logical boundary.

Preference Weights

Each Nest has three preference weights that influence how ANT ranks the models in the nest:

Weight	Effect	Example
Speed	Favors models with lower latency	Set high for real-time chat, autocomplete
Quality	Favors models with higher capability scores	Set high for code review, analysis, reasoning
Cost	Favors models with lower per-token pricing	Set high for high-volume, budget-sensitive workloads

Weights are set on a 0–100 scale. ANT combines them into a composite score for each model in the Nest and picks the winner. You can adjust weights at any time without changing application code.

Example Nest Configurations

fast-coding

S:80 Q:70 C:50

GPT-4o, Claude Sonnet, Gemini Flash. Optimized for quick code generation with good quality. Ideal for AI coding assistants.

budget-chat

S:60 Q:40 C:100

Free-tier and low-cost models only. Great for high-volume chatbots, prototyping, and internal tools where cost matters most.

compliance-us

S:50 Q:80 C:40

Only models from US-hosted providers. Ensures all data stays within approved infrastructure boundaries.

deep-think

S:20 Q:100 C:20

o1, Claude Opus, DeepSeek R1. Maximum reasoning capability for the hardest problems. No speed or cost constraints.

Nests vs Virtual Models

Feature	Virtual Models	Nests
Model selection	All available models	Only the models you selected
Preferences	Preset (auto, fast, best, etc.)	Fully customizable per Nest
Key locking	Not supported	API keys can be locked to a Nest
Best for	Quick start, general-purpose use	Production, compliance, cost control
Usage	`ant:auto`	`nest:your-name`

You can use both. Start with Virtual Models for prototyping, then create Nests when you need fine-grained control for production.