farningcode-api (0.1.0)
Installation
pip install --index-url About this package
Farning Code API
Usage
uv sync
uv run main.py
When main.py runs from a source checkout, it copies the adjacent core source
into API/dev_core on startup. Restart the API after editing core; no rebuild is
needed. Sessions and settings both use this copy and the same config file.
uv sync also installs farningcode-core as a local package dependency.
For deployment, build the core wheel from the repository root with uv build,
then install that wheel alongside the API dependencies. Deployments without the source tree use the installed package.
Configuration is shared through the user config directory (FarningCode/config.json
under Application Support on macOS, LOCALAPPDATA on Windows, or ~/.config on Linux).
Set FARNING_CODE_CONFIG to an explicit config file to use an existing deployment
configuration. Repository config files are no longer selected automatically.
Best accessed through the Farning Code UI.
Settings
GET /settings returns configured model IDs and display names, selected_model,
ignore_folders, and auto_approve_tools. Model connection hosts, endpoint types, and a has_api_key flag are included; API keys are excluded.
PATCH /settings saves any of these preferences to the shared configuration:
{
"selected_model": "your-configured-model-id",
"ignore_folders": [".git", ".venv"],
"auto_approve_tools": ["file_reader"]
}
To add, edit, or remove model connections, include a complete models array in
PATCH /settings, together with selected_model when changing the active ID:
{
"models": [{
"id": "local-model",
"name": "Local model",
"api_host": "http://localhost:1234",
"ai": "openai",
"api_key": ""
}],
"selected_model": "local-model"
}
Model IDs must be unique; hosts must use HTTP(S). Only OpenAI-compatible endpoints
are supported. Omit api_key to preserve a saved key, supply a value to replace it,
or an empty string to clear it. Include original_id when renaming an existing
model to retain its key. Omitted models are removed; use an empty array and an
empty selected_model to remove all connections.
Allowed tools: file_reader, file_writer, file_deleter, execute_code.
An empty tool list requires manual approval for every tool. Invalid updates return
400; persistence failures return 500 and restore the in-memory preferences.
Model selection applies to new sessions, ignored folders to subsequent workspace
scans, and tool approval changes to future tool batches. Already queued approvals
remain unchanged. Run this local API only for trusted users; settings are shared.