Antbase/Documentation

OpenCode Integration

Use ANT as the backend for OpenCode to get intelligent model routing for your coding sessions.

Setup via Custom Provider

OpenCode supports adding custom OpenAI-compatible providers through its settings UI. Open OpenCode settings and add ANT as a custom provider with the following details:

Provider IDant
Display nameANT
Base URLhttps://antbase.ai/v1
API keyant_...Your ANT API key from the dashboard

After saving, select ANT as your provider in OpenCode and choose a model. For coding, we recommend ant:coder or ant:reasoning. All requests will route through ANT, which picks the best underlying model automatically.

Alternative: Environment Variables

You can also configure ANT via environment variables before launching OpenCode:

export OPENAI_API_BASE=https://antbase.ai/v1
export OPENAI_API_KEY=your-ant-api-key
opencode

Configuration File

For a persistent setup, add ANT as a provider in your OpenCode config file (~/.opencode/config.json or opencode.json in your project root):

{
  "provider": "openai-compatible",
  "baseUrl": "https://antbase.ai/v1",
  "apiKey": "${ANT_API_KEY}",
  "model": "ant:coder"
}

The ${ANT_API_KEY} syntax references an environment variable so you don't have to hardcode your key in the config file.

Best Models for Coding

ANT exposes virtual model names that automatically route to the best underlying provider. For coding with OpenCode, reasoning models are the best choice — they think step-by-step through complex logic, catch edge cases, and produce more reliable code.

Coding & Reasoning

ant:coderbest for coding

Optimized for code generation, refactoring, and debugging. Routes to the highest-quality coding models available. Use this as your default model in OpenCode.

ant:reasoningdeep thinking

Routes to thinking-capable models (DeepSeek-R1, Qwen3-Thinking, etc.) that reason through problems step-by-step. Excellent for algorithmic problems, complex debugging, and architecture decisions.

ant:reasoning-premium

Maximum quality reasoning. Only routes to the most capable thinking models (quality score 85+). Use for the hardest problems where correctness matters more than speed.

ant:reasoning-fast

Fast reasoning for quick iterations. Still uses thinking models but favors speed over depth. Good for simpler logic problems and rapid prototyping.

General Purpose

ant:best

Highest quality non-reasoning model. Great for complex tasks that don't need step-by-step thinking — documentation, code review, explanations.

ant:auto

Smart routing based on prompt complexity. Simple questions get fast models, hard questions get powerful ones. Good if you want one model for everything.

ant:balanced

Balance between quality, speed, and cost. Minimum quality score of 60.

ant:fast

Lowest latency. Ideal for inline completions, quick edits, and simple questions.

ant:free

Zero-cost tier using free provider models. Good for experimentation and non-critical tasks.

Specialized

ant:vision

Vision-capable models for analyzing images, screenshots, diagrams, and UI mockups.

ant:creative

Optimized for creative writing, brainstorming, and content generation.

How to Test

Before launching OpenCode, verify that ANT is reachable and your key works:

Health Check

# Verify ANT is reachable
curl -s https://antbase.ai/healthz

Test a Coding Request

curl -X POST https://antbase.ai/v1/chat/completions \
  -H "Authorization: Bearer $ANT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "ant:coder", "messages": [{"role": "user", "content": "Write a hello world in Rust"}], "max_tokens": 200}'

You should get a JSON response with a Rust hello world program. Check the X-Antbase-Model response header to see which model ANT selected.

Troubleshooting

  • "Connection refused"

    Check that the Base URL is set correctly and includes the /v1 suffix: https://antbase.ai/v1

  • "401 Unauthorized"

    Verify your API key is correct and starts with ant_. You can find your API key in the ANT dashboard under API Keys.

  • "Model not found"

    Use virtual model names like ant:auto, ant:best, or ant:fast. Raw provider model names (e.g. gpt-4o) are also supported if your plan includes direct routing.