Buy the Harbor One appliance. Use it.

Harbor One is a dedicated AI coding appliance from Enba. It runs supported coding models locally on hardware you own and exposes compatible API interfaces for supported coding-agent workflows.

No. Harbor One is an on-premises appliance that runs on your own network. Enba may offer a hosted API evaluation so you can test the workflow before buying hardware, but the Harbor One product itself is customer-owned hardware.

For local inference requests sent to Harbor One, source code, prompts, context, and model outputs are processed on your own infrastructure rather than sent to a cloud model provider.

You can use Harbor One as much as its finite hardware capacity allows. Model speed, context length, concurrent workloads, electricity, and agent behavior still determine practical capacity.

Enba supports selected open coding models that fit the appliance and perform well on practical coding-agent workloads. Because the model ecosystem changes quickly, Harbor One is designed around supported model profiles rather than a permanent commitment to one model release.

Not as part of the standard Harbor One product. Harbor One is designed to work without Enba ingesting your repositories. Custom tuning or private adaptation may be offered separately in the future.

Harbor One is designed to expose an OpenAI-compatible endpoint for supported Codex-oriented workflows. Compatibility depends on the exact client, version, model profile, and workflow.

Harbor One is designed to expose an Anthropic-compatible endpoint for supported Claude Code-oriented workflows. Compatibility depends on the exact client, version, model profile, and workflow.

Not always. The strongest frontier cloud models may outperform local models on the hardest tasks. Harbor One is for developers and teams that value the combination of strong coding capability, local control, predictable economics, and freedom from Enba per-token pricing. The practical question is how much of your real coding work can be handled well locally and whether owning that capacity is valuable to you.

You can. Harbor One is for people who do not want to spend their time selecting components, maintaining drivers and inference servers, testing quantization formats, tuning context settings, adapting agent endpoints, and retesting every new model release. Enba turns that ongoing integration work into a supported appliance.

$9,499 for the appliance, plus the $100/month Enba Support Plan for updates, diagnostics, supported model profiles, and Enba software support. Local inference has no Enba per-token or per-seat fee.

Yes. The 30-day API evaluation is $500, and the fee can be applied toward a qualifying Harbor One purchase made within 60 days.

That depends on the model profile, context length, agent behavior, and how heavily people use it. Harbor One provides finite dedicated capacity rather than elastic cloud capacity.

Local model inference itself does not require a cloud model provider. Network requirements for setup, updates, support, and diagnostics may vary by deployment. Talk to Enba about restricted-network or offline requirements.

The Enba Support Plan includes supported model-profile updates. Enba evaluates new models and can add profiles when they materially improve the experience on Harbor One hardware.