Open Source • European AI • Offline First

Sovereign AI Coding Agent for Developers Who Need Control

EuroCoder gives you Cursor / Claude Code-like capabilities without routing your source code through US cloud providers. Use Mistral for high quality or Ollama for 100% local execution.

No telemetry API keys stored locally Planner + coder architecture Docker + Git + Shell tools

CLI in Action

euro-coder:> ask check my running docker containers

Agent: Here are your currently running containers:

CONTAINER ID   IMAGE          STATUS         PORTS                    NAMES
a1b2c3d4e5f6   postgres:16    Up 3 hours     0.0.0.0:5432->5432/tcp   local-db
f6e5d4c3b2a1   redis:7        Up 3 hours     0.0.0.0:6379->6379/tcp   cache

Why EuroCoder Exists

Many AI coding assistants depend on US cloud infrastructure. EuroCoder is built for teams and developers who care about digital sovereignty, data boundaries, and regulatory control.

Privacy by Design

  • Your machine, your data
  • Zero telemetry and no tracking
  • Local config with owner-only permissions
  • Offline mode with Ollama

Agent Capabilities

  • Project tree, file search, and content grep
  • Read/write files and directory management
  • Shell execution for git, docker, npm, curl, and more
  • Project scaffolding and code modifications

Hybrid Planner / Coder Architecture

EuroCoder routes simple prompts directly and uses a two-model strategy for complex work. The planner explores your project and creates an execution plan; the coder executes with file-level changes.

Supports auto mode, single-model mode, and custom planner/coder pairings.

Dual Provider Support

Provider How it works API key Best for
Mistral (default) Cloud inference on European servers Yes Best quality and low latency
Ollama Runs models entirely on your machine No Maximum sovereignty and offline mode

Quick Start

git clone https://codeberg.org/euro-coder/eurocoder.git
cd eurocoder
./mvnw clean package -DskipTests
java -jar target/sovereign-agent-0.1.0-SNAPSHOT.jar

Requires Java 21+ and either a Mistral API key or local Ollama setup.

First Run Setup

  1. Choose provider: Mistral Cloud or Ollama Local
  2. Enter API key (if Mistral)
  3. Select planner/coder model pairing

Core Commands

  • ask <prompt> for normal agent routing
  • plan <prompt> to force planner + coder mode
  • code <description> for direct fast generation
  • provider ollama to switch to local mode

Technology Stack

Roadmap

  • MCP (Model Context Protocol) integration
  • Local RAG for codebase-aware retrieval
  • VS Code and IntelliJ integrations
  • More providers and local GGUF options
  • Streaming responses and session persistence

License

Released under EUPL-1.2 (European Union Public License), valid across EU languages and compatible with major open-source licenses.

Contribute

Contributions are welcome. Open an issue first to discuss changes, then submit a pull request.

Browse Issues

FAQ

Can I run EuroCoder fully offline?

Yes. Use Ollama and local models for zero network calls.

Where is configuration stored?

In ~/.eurocoder/config.json, with owner-only file permissions. API keys can also be passed via MISTRAL_API_KEY.

What hardware do I need for local models?

8 GB RAM can run smaller models; 16 GB+ is better for larger model pairs. Apple Silicon devices are especially effective because of unified memory.

Build AI workflows on your terms

Deploy EuroCoder today and keep your coding assistant aligned with privacy, sovereignty, and open source values.

Get Started on Codeberg