Bring Your Coding Agent, Keep Your Code in Europe

Coding agents are becoming standard tooling for developers. Most coding agents ship with a default LLM backend that routes your code, your prompts, and your intellectual property through infrastructure you don't control.

If you're building for a European enterprise, a regulated industry, or any organisation that treats data sovereignty as a priority then that is a blocker.

Some teams have chosen to self-host an open weight model and handing the management themselves. But then you are managing quantization, context window limits, and GPU scheduling instead of shipping software. The trade-off is clear: control at the cost of velocity.

We built evroc Think for teams that want both. Think gives you dedicated or shared GPU-backed inference inside Europe with an OpenAI-compatible API.

The missing piece was the agent wiring: getting your preferred coding agent to actually talk to Think instead of its default backend in a developer friendly way.

That is what evroc think code does.

The Agent Side of Sovereign Inference

evroc Think has always been an inference service. You provision a model instance or use a shared one, you get an endpoint and an API key, and you call it from your application.

For these cases, you can create an API keys, view usage your usage per model and choose between shared or dedicated models directly in the evroc console.

But you probably don't write raw API calls when you code but instead use one of the popular coding harnesses like OpenCode or Pi.

These tools allow you to customize the remote backend for which LLM you are using. It can require copying endpoint URLs, API keys, model configs and formats. This can be easy when doing it just for yourself but extremely hard to scale for a team.

To make this process more efficient, we added evroc think code feature to the evroc CLI.

evroc think code collapses the setup up steps into a single command:

evroc think code [coding-agent-name]

The CLI reads your Think credentials from ~/.evroc/config.yaml, discovers your available models, and configures the agent to use them. No manual base URLs. No copy-pasted API keys into agent config files. One command, your agent, on European hosted infrastructure.

What You Can Connect

We currently offer 10 different coding harnesses that allow you to route inference Think model running in Europe:

• OpenCode
• Hermes
• Pi
• Aider
• Claude Code
• Codex CLI
• Crush
• Goose
• OpenClaw
• QwenCode

You connect any of them in one command to keep the reasoning, tool use, and editing strategy of the harness.

Only the LLM backend changes, and that change is what keeps your prompts and your code within a EU sovereign host.

Making It Permanent

Running evroc think code [agent] once configures the agent for the current session.

If you want defaults that persist across terminal sessions and projects, there's an init command:

evroc think code init

This writes a profile to ~/.evroc/think-code.yaml with your default model, preferred agent, and any project-specific overrides.

After init, just running evroc think code (with no agent argument) launches your default harness against your default Think model.

Getting Started

If you already have a Think model running:

# See available models
evroc think model list

# Connect Aider to your default Think model
evroc think code aider

# Or set up a permanent profile
evroc think code init

If you don't have a Think model yet, the fastest path is a shared model:

evroc think model create shared-llama --type shared --model llama-3.3-70b

Shared models start instantly and scale to zero when idle. For teams with steady workloads, dedicated instances give you predictable latency and full GPU isolation.

Provision them with the same CLI:

evroc think model create dedicated-gpu \
  --type dedicated \
  --model llama-3.3-70b \
  --gpu-profile nvidia-a100

Wrapping Up

Ready to connect your agent?

For the first 100 signups, we are offering €300 credits to help you get a started building on evroc and using evroc Think.

To claim your credits, sign up here.

For more information about evroc Think, see the evroc Think docs.