OpenCode with PrivatAI: a coding agent that keeps your code in the EU

OpenCode with PrivatAI

OpenCode is a coding agent for the terminal: open source, provider-agnostic, running in your project directory. It reads files, writes patches, runs commands and works through tasks on its own.

That makes data protection concrete. A coding agent doesn't see a handful of prompts — it sees your codebase. Configuration, customer names in test fixtures, comments about known weaknesses, internal interfaces. With a US provider, all of that leaves the EU. Not occasionally, but on every request, all day long.

OpenCode doesn't pick a provider for you. Any OpenAI-compatible endpoint can be wired in. PrivatAI is one.

flowchart LR
    A[OpenCode<br/>your terminal] -->|default| B[US provider]
    A -->|with PrivatAI| C[PrivatAI<br/>GER/FR]

Where most endpoints fall over

A coding agent needs tool calling. Without it, it cannot open a file or run a command — it can only talk. Plenty of endpoints call themselves "OpenAI-compatible" and then quietly drop the tools field. The agent starts, looks normal, and does nothing.

PrivatAI passes tools and tool_choice through to the model and returns tool_calls — streaming and non-streaming alike. OpenCode runs fully, not as a glorified chat window.

Prerequisites

A PrivatAI account on an active plan (sign up here)
An API key from your dashboard (starts with privat_)
A terminal — the Terminal app on macOS, any shell on Linux, WSL on Windows (see step 1)

Step 1: Install OpenCode

macOS

brew install opencode

No Homebrew? The Linux command below works on macOS too.

Linux

curl -fsSL https://opencode.ai/install | bash

Windows

OpenCode recommends WSL on Windows, a Linux environment built into Windows. In PowerShell:

wsl --install

After the restart, open "Ubuntu" from the Start menu and run the Linux command there. Do every following step in that Ubuntu window.

Step 2: Save your API key

Create an API key in the dashboard and copy it — it's shown only once. Replace privat_your_key with your key and run:

macOS

echo 'export PRIVATAI_API_KEY="privat_your_key"' >> ~/.zshrc

Linux and Windows (Ubuntu window)

echo 'export PRIVATAI_API_KEY="privat_your_key"' >> ~/.bashrc

Then open a new terminal window. Windows that were already open don't have the key yet, and OpenCode will answer every request with Unauthorized.

Step 3: Add PrivatAI to OpenCode

Copy the whole block below into your terminal and press Enter to set up the PrivatAI models:

mkdir -p ~/.config/opencode
if [ -e ~/.config/opencode/opencode.json ]; then
  echo "Stopped: ~/.config/opencode/opencode.json already exists and was left unchanged."
else
  cat > ~/.config/opencode/opencode.json <<'EOF'
{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "privatai": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "PrivatAI (EU)",
      "options": {
        "baseURL": "https://privatai.com/api/v1",
        "apiKey": "{env:PRIVATAI_API_KEY}"
      },
      "models": {
        "gpt-oss-120b": {
          "name": "gpt-oss-120b",
          "limit": { "context": 128000, "output": 8192 }
        },
        "glm-5.2": {
          "name": "GLM-5.2",
          "limit": { "context": 128000, "output": 16384 }
        }
      }
    }
  },
  "model": "privatai/gpt-oss-120b"
}
EOF
  echo "Done: PrivatAI models added to OpenCode."
fi

Step 4: Pick the model

gpt-oss-120b is preselected. To switch, type /models inside OpenCode, or choose at startup:

opencode --model privatai/glm-5.2

gpt-oss-120b is included in every plan; glm-5.2 requires Professional.

Step 5: Check that it works

First the key, straight against the API:

curl -X POST https://privatai.com/api/v1/chat/completions \
  -H "Authorization: Bearer $PRIVATAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "gpt-oss-120b", "messages": [{"role": "user", "content": "Say hello"}]}'

Then OpenCode, inside one of your projects:

cd your-project
opencode run "What language and framework does this project use?"

If the answer describes your project rather than staying generic, the agent read your files — tool calling works.

Run `/init` once

Start opencode in the project and type /init. OpenCode analyses the repository and writes an AGENTS.md: structure, conventions, build commands. Commit that file. It saves you clarifying questions — and tokens — on every task afterwards.

Keeping an eye on cost

An agent uses more than a chat does — it reads, tries, corrects. To see what your sessions used:

opencode stats

That's it

Three blocks pasted into a terminal. OpenCode stays what it is: your agent, your terminal, your code on your disk. What changes is where the requests go — France instead of Virginia. And what your agent reads along the way doesn't become training data.

Questions or feedback? Write to us at info@privatai.com.

OpenCode with PrivatAI: a coding agent that keeps your code in the EU