One AI backend.
Every editor.
ANT speaks the Agent Client Protocol — the emerging standard for connecting code editors to AI agents. Zed, JetBrains, Kiro, and any ACP-compatible editor can use ANT as their AI backend. 20,000+ models, intelligent routing, one configuration.
What is the Agent Client Protocol?
ACP is like LSP, but for AI agents. Just as the Language Server Protocol standardized how editors talk to language servers (enabling Go, Rust, TypeScript support in every editor), ACP standardizes how editors talk to AI coding agents.
Before ACP, every AI coding tool needed a custom integration for every editor — Copilot for VS Code, Cursor's built-in AI, JetBrains AI Assistant. With ACP, any agent works with any editor. And ANT makes that agent intelligent — routing every request to the best model from 20,000+ options.
ACP + ANT = Any editor connects to one AI backend that has access to every model in the ecosystem, with automatic model selection, failover, and cost optimization. No vendor lock-in, no single-model limitations.
Learn more about ACP at agentclientprotocol.com →Works with every ACP editor
High-performance editor with native ACP support
SupportedIntelliJ, PyCharm, WebStorm, GoLand — full ACP integration
SupportedAWS-backed AI IDE with ACP agent protocol
SupportedVia ACP extensions and compatible agents (Cline, Continue)
Via extensionsTerminal-based with ACP plugin ecosystem
CommunityAny editor implementing the Agent Client Protocol
CompatibleHow ANT + ACP works
Editor sends prompt via ACP
You ask your editor to refactor code, explain a function, or generate tests. The editor sends the request using ACP's JSON-RPC protocol.
ANT classifies and routes
ANT analyzes the prompt — task type, complexity, required capabilities. It scores 20,000+ models and selects the optimal one in milliseconds.
Best model responds
The selected model (Claude Opus for hard refactors, GPT-4.1 mini for simple completions, Llama for free experiments) streams the response back to your editor.
Fixed model agent vs ANT-powered agent
Most ACP agents hardcode one model. ANT makes any agent smart.
| Feature | Fixed Model Agent | ANT-Powered Agent |
|---|---|---|
| Models available | 1 model (hardcoded) | 20,000+ with intelligent selection |
| Model selection | Manual configuration | Automatic per task |
| Failover | None — if provider is down, agent breaks | Automatic — routes to next best provider |
| Cost | Same price for everything | 30-50% savings via smart downrouting |
| Free models | Not available | Llama, Gemma, Qwen via ant:free |
| New models | Manual reconfiguration | Automatic — new models available instantly |
| Code change | N/A | One config: set base_url to antbase.ai/v1 |
Configure your ACP agent to use ANT
{
"provider": "openai-compatible",
"baseUrl": "https://antbase.ai/v1",
"apiKey": "your-ant-api-key",
"model": "ant:coder",
"capabilities": {
"chat": true,
"vision": true,
"tools": true,
"streaming": true
}
}ANT's OpenAI-compatible API works with any ACP agent that uses the OpenAI SDK under the hood — which is most of them. Set the base URL, provide your API key, and choose a virtual model.
Virtual models for ACP agents
Best code model for each task — Claude, GPT-5, DeepSeek Coder. The default for ACP agents.
Chain-of-thought for architecture decisions, complex debugging, algorithm design.
Sub-200ms for inline completions and quick suggestions. No interruption to your flow.
Zero cost with Llama, Gemma, Qwen. Perfect for experimentation and learning.
Maximum quality for critical code reviews and production decisions.
Let ANT decide everything — simple edits get fast models, hard problems get frontier.
Make your ACP agent intelligent
One config. 20,000+ models. Every ACP-compatible editor. $5 free credits to start.