Settings

Blueprint AI exposes two settings sections — one per-developer, one per-project.

Section	Stored in	Committed to source control?
Editor Preferences → Blueprint AI	`EditorPerProjectUserSettings.ini`	No — gitignored by convention
Project Settings → Blueprint AI	`DefaultGame.ini`	Yes — shared across the team

The split is intentional: anything that describes this developer’s machine (CLI paths, account locker, agent ordering) lives in Editor Preferences; anything that describes the project itself (working directories, ignore patterns) lives in Project Settings.

Editor Preferences

Editor → Editor Preferences → Blueprint AI

General

`Agents`

An ordered list of AI coding assistants to install the Blueprint AI skill for. Each entry causes setup.py to copy the bundled skill file to the assistant’s skills directory on the next editor launch.

Value	Skill installs to	Bundled in-editor chat panel
`Claude`	`.claude/skills/unreal/`	✅

Claude is the supported assistant — both for skill auto-install and for the bundled chat panel. Cursor, GitHub Copilot, Windsurf, and OpenAI Codex ship next. For now, drive those tools from their own UI or use the Command Line agent as the in-panel escape hatch. See AI Service Setup for the full walkthrough.

Claude

Visible only when Claude is in the Agents list.

`Claude CLI Path`

Explicit path to the claude executable. Leave empty for auto-detection — the plugin checks PATH, then the standard install locations (%LOCALAPPDATA%\anthropic\ on Windows, ~/.local/bin/ and /usr/local/bin/ on POSIX, the official npm install location, etc.).

Set this when you have a non-standard install, or when you want to pin to a specific version while another lives on PATH.

`Save Claude accounts after login`

When on (default), successful Claude logins are copied into ~/.claude/BlueprintAI/<email>/ so you can switch between accounts without redoing OAuth. See Multi-Account Login for the full flow.

Turn off if you don’t want past logins persisted on disk — the panel still works, you just sign in fresh each time you switch accounts.

Telemetry

`Send anonymous error reports`

Off by default. When the user explicitly opts in (either by toggling this checkbox or by accepting the pre-checked default on the login screen), the plugin reports anonymous crashes and errors via Sentry.

What’s sent:

Exception type, stack trace, plugin version, engine version, OS
Opaque random install ID (so events can be grouped without identifying you)
Editor selection (current sidebar / focus only — never asset content)

What’s not sent:

Source code, file contents, chat messages, attachment contents
File paths beyond the plugin directory
Account identifiers, emails, API keys

Full disclosure on the Privacy & Telemetry page. Flip the checkbox to revoke consent — telemetry stops immediately and the install ID is rotated on next opt-in.

Project Settings

Edit → Project Settings → Blueprint AI

These ship in DefaultGame.ini, so committing them with the project applies the same setup to every teammate.

Autocomplete

`Autocomplete Working Directory`

Root directory the @ autocomplete walks. Relative to the project root. Leave empty to inherit Agent Working Directory — then the @ menu and the agent always agree on what “the project” is.

Setting this explicitly is useful when your project sits inside a larger repo (e.g., engine + game + tools all in one tree) and you want @ to stay focused on the game subdirectory.

`Autocomplete Ignore`

A gitignore-style pattern list pruning entries from @ autocomplete. Blank lines and # comments are skipped.

Pattern semantics:

Lines without / match the basename at any depth — .git/, *.uasset, node_modules/
Lines with / are anchored to the working directory — Source/Generated/, Content/Fab/**
Trailing / limits a pattern to directories
Leading ! re-includes a previously ignored path — useful for “ignore all of X/ except X/important.md”

Reasonable defaults to start with:

# Build output
Binaries/
Intermediate/
Saved/
DerivedDataCache/

# Asset files — use Asset Registry instead of raw .uasset
*.uasset
*.umap

# Engine source if you have the source build
Engine/

Changes take effect on the next @ query — no editor restart.

Agent

`Agent Working Directory`

Working directory every panel-spawned process runs in:

The Claude sidecar (the OAuth-backed agent that powers chat)
The claude remote-control session the panel attaches to
The interactive agent CLI in Terminal mode when using a non-Claude agent
The platform shell launched by the Command Line agent

Relative to the project root. Empty = the project root itself.

Set this when:

Your .uproject lives in a subdirectory (Game/MyGame.uproject in a monorepo) and you want the agent’s view of “the project” to start at the subdirectory
You have a build / tools subdirectory you want the agent rooted in (e.g., to keep its pwd near build scripts)

The setting is always editable — it applies even when you haven’t configured an agent yet, since the Command Line terminal is reachable from the login screen before sign-in.