kolega-code 0.1.0

Local-first AI coding agent for the terminal

kolega-code

Kolega Code is a local-first AI coding agent for the terminal.

The package owns the kolega_code import namespace and provides the kolega-code command.

Install

Install with the public installer:

curl -fsSL https://kolega.dev/install-kolega-code | sh

Or install directly from PyPI with uv:

uv tool install kolega-code

Verify the command is available:

kolega-code --version

Upgrade or uninstall:

uv tool upgrade kolega-code
uv tool uninstall kolega-code

Run the Textual UI and open the Settings tab to select Moonshot Kimi K2.6 or DeepSeek V4 Pro and save your API key:

kolega-code .

In the Textual UI, press Shift+Tab to switch between build mode and planning mode. Planning mode uses a standalone read-only planning agent; when it submits a complete plan, choose whether to implement it or keep discussing the plan.

All CLI sessions use the CLI-specific coding-agent prompt, including resumed sessions. Launching the UI starts a fresh thread by default. Resume an existing thread explicitly:

kolega-code . --resume
kolega-code . --resume <thread-or-session-id>

You can also set MOONSHOT_API_KEY, DEEPSEEK_API_KEY, or keep using env/flag based configuration for non-UI commands:

kolega-code ask "summarize this repository" --project .
kolega-code ask "summarize this repository" --project . --provider deepseek --model deepseek-v4-pro
kolega-code sessions list --project .
kolega-code doctor --project .

The Settings UI supports Moonshot kimi-k2.6 and DeepSeek deepseek-v4-pro. A saved UI selection is used for all agent model roles and API keys are stored in the local CLI settings file with restrictive permissions. Existing environment and model/provider flag overrides continue to work. Local session state is stored under the platform state directory unless KOLEGA_CODE_STATE_DIR is set.

From source

git clone https://github.com/kolega-ai/kolega-code.git
cd kolega-code
uv sync --extra dev
uv run kolega-code --version

Tests

Fast tests run by default:

./run_tests.sh

Some slow and integration tests require real provider credentials. To run them locally, create an ignored .env file from the example and fill only the keys you need:

cp .env.example .env
./run_tests.sh --all

The test runner loads .env through pytest and keeps existing shell environment variables higher priority than values in the file. You can pass additional pytest arguments through the wrapper:

./run_tests.sh kolega_code/agent/tests/llm/test_client.py -ra