Five minutes to your first Scout edit.
Pearl is installed and you're here. Below is the five-minute path: sign in, pick a mode, ship one small thing, and you'll know if Pearl fits how you work.
Four steps. None of them is "watch a video."
Sign in
Open Pearl. Click Sign in to Pearl Scout in the chat panel.
A browser tab takes you through Pearl SSO and hands you back to the IDE via
the pearl:// protocol handler. You're in.
Open a real project
File → Open Folder, point at a repo you know well. Scout works better when you pick something you can sanity-check the answers against. (You can also use multi-root to open several projects at once.)
Pick the right mode
The toolbar above the chat input has a mode selector. Start with
Ask — it's read-only, so you can poke around without
Scout writing anything. "What does UserService.php do?" is a fine
first question.
Try one small Agent edit
Switch to Agent. Pick a small task — "add a docstring to this function" or "extract this constant." The edit lands as an inline diff with per-hunk Keep/Undo. If you don't like it, one Ctrl+Z reverts the whole thing.
Eight things that make Pearl feel fast.
Attach the active file
Type #file in the chat input. Or #_f for the shortcut.
Attach the selection
Highlight code, type #selection. Or #_s.
Attach a specific file by path
#file:src/users/UserService.php. Tab-complete works.
Switch agents mid-chat
Type @Pearl-Ask, @Pearl-Plan, or @Pearl-Agent to route one message without changing modes.
Click file references
Any backtick path in chat is a link. UserService.php:42 jumps to line 42.
Undo a whole response
One Ctrl+Z reverts every hunk in an Agent response at once.
Compact when context fills up
Click the green/orange/red ring in the chat input to summarize older turns and free the budget.
New chat with summary
"New chat with summary…" hands off your current context cleanly into a fresh session.
About .pearl/memory/project.md
Scout keeps per-project memory in a plain markdown file at
.pearl/memory/project.md. It's yours — read it, edit it, commit it
if you want.
A memory chip in the chat input shows the current entry count; clicking it opens the file. Scout reads memory at the start of relevant turns and adds entries when it learns something durable about your project.
If you don't want memory: delete the file, or open it and remove entries. Nothing automatic puts it back.
The honest fine print
- Scout needs a network. Chat doesn't work offline.
- A Pearl account is required — including for Codex, Claude Code, and Copilot providers.
- Your provider API keys (BYO) stay in the OS keychain. Pearl never sees them in plaintext beyond the handoff.
- We don't train on your prompts. Scout reads code at request time as context — not as training data.
If today went well…
Try Plan mode
For multi-file refactors. Plan reads the workspace, emits a phased checklist, and you Run each phase in Agent mode.
Bring your own keys
If you already have Anthropic / OpenAI / Google keys, plug them in via the provider picker. Your quota, your bill.
Read the docs
Keybindings, advanced settings, troubleshooting. The bits you reach for when something doesn't behave.
File feedback
Bugs, feature requests, sharp edges. Tell us — it lands in the same queue we work from.