pi-ooc

Sometimes you want to ask pi a side question without dragging that detour back into your main conversation, or did you ever wonder what your agent is doing, but you don't want to interrupt it? Call /ooc what are you doing?, and its clone will tell you without hurting the progress of the main agent.

pi-ooc adds /ooc (think "out of context"), which opens an isolated side-agent with the full current session context, shows the result in an overlay, and keeps that whole exchange out of your main session history.

Think of it as: "use everything we know so far, but don't make this part of the main thread."

pi-ooc is intentionally minimal by design. If you want a more full-featured alternative, have a look at pi-btw.

Install

Option 1: Install from npm (recommended)

pi install npm:@mjakl/pi-ooc

Option 2: Install via git

pi install git:github.com/mjakl/pi-ooc

Option 3: Install local package

pi install ./

Usage

/ooc What assumptions have we made so far?
/ooc Give me three alternative designs for this refactor.
/ooc Inspect the repo and tell me where the auth flow starts.
/ooc Challenge the current plan and tell me what we're missing.

A nice extra is to /ooc what are you doing? while the main agent is working on something that looks off.

What happens

When you run /ooc ...:

1. pi waits for the current agent to become idle if needed
2. a separate side-agent session is started
3. that side-agent gets the full current session context
4. it uses the current model and thinking level
5. it is told it is a read-only agent that only answers the question, briefly
6. it keeps pi's built-in tools, but edit and write calls are blocked
7. its output is streamed into a TUI overlay
8. when you close it, nothing from that exchange is appended to your main session

Why use it?

/ooc is useful when you want to:

• ask a side question without cluttering the main thread
• get a second opinion based on the current context
• inspect the repo or run tools without turning that detour into part of the main conversation
• challenge the current plan
• explore alternatives before committing to a direction
• do a quick isolated investigation and then return to your main flow

Important behavior

This is the part that matters most:

• the /ooc conversation is not added to your current session history
• the result is shown only in the overlay
• the side-agent has the same context as your current session
• the side-agent is instructed to be read-only and to answer short and to the point
• edit and write calls are blocked and come back as an error the side-agent can read
• extension-provided tools are intentionally not loaded; an extension's background work could outlive the short-lived side session
• bash is still bash; the read-only rule is an instruction, not a sandbox, so a command that modifies something has real effects
• the side-agent adds nothing to the tools and nothing to the system prompt, so its request prefix stays as close to your main session's as it can be and a provider can reuse the cached prefix; that is why the read-only instructions ride at the end of the question
• full reuse is not guaranteed: extension-registered tools are part of your main session's prefix but not the side-agent's, and providers that key their cache on the session id (such as openai-codex) see a different id for the side session

So /ooc is isolated from your conversation history, but not from your working directory.

Closing the overlay

If the side-agent is still running and you press Esc or q, pi-ooc will not close immediately.

Instead it shows a confirmation modal:

• press Esc or q again to abort and close
• press any other key to keep it running and continue reading

Once the side-agent is finished, Esc or q closes the overlay normally.

Keys inside the overlay

• Esc or q - close
• ↑ / ↓ - scroll
• g / G - jump to top / bottom

In one sentence

/ooc gives you a fully context-aware side-agent in an overlay, without polluting your main session history.

License

MIT