Meet Harbor One.
A dedicated local AI system for coding-agent workloads. Harbor One combines high-performance GPU hardware, a configured inference stack, supported coding-model profiles, compatible API endpoints, and ongoing support in one appliance you own.
Not a generic workstation. Not another cloud subscription.
Harbor One is built around a simple idea: serious local AI should be deployable as a product, not assembled as an ongoing infrastructure experiment. Enba handles the integration work between hardware, drivers, inference software, model formats, quantization, context configuration, API behavior, diagnostics, and updates.
You connect the system to your network and point supported coding tools at it. Enba maintains the supported stack.
Everything required for practical local AI coding.
Purpose-built hardware
A high-performance GPU system selected for running strong local coding models.
Configured inference stack
The serving software is installed, tested, and tuned so your team does not need to become expert in drivers, quantization formats, context settings, or inference servers.
Supported model profiles
Enba selects and configures coding models for practical quality, speed, context handling, hardware fit, and agentic tool use.
Coding-agent compatibility
Familiar OpenAI-compatible and Anthropic-compatible interfaces for supported workflows, including tools in the Codex and Claude Code ecosystems.
Local management and diagnostics
Operate the appliance on your network, with supported diagnostics, model-profile updates, and Enba support available through the Support Plan.
Built around a durable appliance and workflow.
The best open coding model for Harbor One today may not be the best one six months from now. Enba continuously evaluates new model candidates and can add supported profiles when they materially improve quality, speed, context behavior, or compatibility on Harbor One hardware.
Harbor One is built around a durable appliance and workflow, not a single model name.
Developers and teams who want local AI without the ops burden.
- Individual developers who use AI coding heavily and dislike usage metering.
- Small engineering teams that want to share dedicated local AI capacity.
- Companies that prefer source code and project context to be processed on infrastructure they control.
- Teams that want less dependence on cloud AI availability, pricing, and policy changes.
- Technical organizations that want local AI without maintaining the entire stack themselves.
Local models. Familiar coding-agent workflows.
Harbor One is designed around compatible API interfaces, so developers can use supported coding agents against local Enba inference without adopting a proprietary editor.
- Claude CodeAnthropic-compatible endpoint for supported Claude Code-oriented workflows.
- CodexOpenAI-compatible endpoint for supported Codex-oriented workflows.
Compatibility is more than an API URL
Different coding agents expect different message formats, tool schemas, streaming behavior, context patterns, and stop conditions. Enba tests complete model-plus-agent configurations so support can be defined around real workflows rather than blanket compatibility claims.
Compatibility depends on the exact client, model profile, endpoint, and workflow. Enba maintains supported configurations around combinations that have been tested together on Harbor One.
# ~/.claude/settings.json
{
"env": {
"ANTHROPIC_BASE_URL": "https://harbor-one.internal"
}
}export OPENAI_BASE_URL="https://harbor-one.internal/v1"
export OPENAI_API_KEY="harbor-team-key"
codex "refactor the billing module"Harbor One exposes familiar OpenAI-compatible and Anthropic-compatible interfaces for supported coding-agent workflows — so your developers keep working in the tools they already know while routing supported workloads to local inference.
See Harbor One on your network.
Start with a 30-day API evaluation, or talk with Enba about deploying Harbor One for your team.