Pi

Model choice

You can leave the model unset in Farhelm. Pi then uses its configured provider and model; Farhelm does not add --provider or --model. Choosing a model explicitly keeps Farhelm’s OpenRouter launch arguments for that choice.

YOLO is the only permission mode

Pi does not ask for permission before running shell commands or editing files, so Farhelm shows YOLO as its only permission option. Pi’s --approve flag instead controls whether Pi trusts settings and extensions supplied by the project you opened; it does not enable approval prompts for the agent’s actions. This was verified against Pi 0.85.1.

The structured launch composer offers workspace trust separately from YOLO. Choosing true adds --approve for that launch; choosing false adds --no-approve. The last explicit choice from a successful user launch becomes the next new dialog’s default. Farhelm does not write Pi’s persistent trust settings.

Resume needs a saved conversation

Pi saves conversations automatically; there is no separate Save step. A new conversation’s file is written after the first assistant message. Send a prompt and let Pi respond; Farhelm can then offer Resume once Pi reports the saved conversation.

Typing /new in Pi starts a new conversation. Until that conversation is saved, Farhelm stops offering Resume for the old one: restarting should not take you back to a conversation you already left.

Farhelm checks the saved file before resuming. If it is missing or belongs to a different conversation, Farhelm refuses Resume and asks you to choose a fresh launch instead. This matters because Pi itself can silently start a new conversation when given a missing session file.

The helper Farhelm uses to track Pi’s active conversation is supplied at launch and is not saved with the conversation. Reopening a Pi conversation outside Farhelm does not require Farhelm to be installed.