You're in. Now what?

The homepage is the Askinterface — type a question in plain English. Right now your brain is mostly empty, so it won't have much to say. That's fine. We're about to feed it.

The left sidebar has the rest of the app:

• Browse — every entity in your brain, filterable by type
• Graph — force-directed view of how entities connect
• Goals — durable multi-week objectives
• Rules — durable instructions that auto-sync to your AI tools
• Skills — Claude Code skills indexed for recommendation
• Optimize — brain health: dedupe, audit, decay, contradictions

The fastest way to feel Roushi's value is to point it at a project you already have locally. It auto-extracts entities (tech, vendors, decisions, patterns, lessons) using an LLM and builds the graph for you.

# Clone your deployed instance locally if you haven't yet
git clone https://github.com/YOUR_USER/roushi
cd roushi
pnpm install
vercel env pull .env.local       # pulls DATABASE_URL etc. from Vercel
pnpm db:setup                    # one-time table + index creation

# Now ingest any project repo you have
pnpm roushi add-project ~/projects/your-project
pnpm roushi add-project ~/projects/another-one

Each add-projectrun takes ~30 seconds. It reads the project's README.md, CLAUDE.md,package.json, and a few other key files, then asks an LLM to extract typed sub-entities (tech, vendor, decision, pattern, lesson). Edges between them get built automatically.

Go back to the homepage. Now ask:

• "What tech do my projects have in common?"
• "Which projects use Postgres?"
• "What decisions have I made about authentication?"
• "Are any of my projects vulnerable to the same lesson I learned in [project name]?"

The answer renders with citations — each [slug]chip links to the entity it's referencing. Below the answer: a gap analysis— what the brain still doesn't know, so you can fill it in next.

Rules are durable instructions that Roushi auto-syncs into your AI coding tools — Claude Code memory dir, Copilot's .github/copilot-instructions.md, Cursor's .cursorrules, Windsurf's .windsurfrules. Once written, every future AI session follows the rule automatically.

Create a file at content/rules/my-first-rule.md:

---
slug: my-first-rule
type: rule
name: A convention I want every AI session to follow
applies_to: portfolio
priority: normal
---

## The rule

Whenever you generate a new API route, validate the request body
with Zod before any database access. Use the Server Action pattern
`{ ok: false, message }` for errors — never throw.

## Why

We've had bugs where unvalidated input flowed into queries.
Catching it at the validation boundary keeps downstream code simple.

Then ingest it:

pnpm roushi ingest content/rules/my-first-rule.md

The rule is now in your brain and will auto-sync into any workspace via the SessionStart hook (Claude Code) or the VS Code extension.

Your repo includes a daily GitHub Action that opens a PR when upstream Roushi has new commits. GitHub disables Actions on new repos by default — enable in one click:

1. Open your repo on GitHub
2. Click the Actions tab
3. Click "I understand my workflows, go ahead and enable them"

Now you'll get a PR titled "Sync from upstream (roushi-template)" whenever there are new commits. Review the diff, click Merge, Vercel redeploys. No more drifting from upstream.

Don't want auto-updates? Skip this step. You can also pull changes manually any time by opening the GitHub compare URL github.com/YOUR_USER/YOUR_REPO/compare/main...samwsimpson:roushi-template:main — creates a PR with the diff for review.

This is where Roushi stops being "a thing you check" and becomes a thing that runs while you code. Pick your editor:

claude mcp add -s user roushi-remote \
  https://your-instance.vercel.app/api/mcp \
  --transport http \
  --header "Authorization: Bearer $YOUR_TOKEN"

code --install-extension kumokodo.roushi
# Then: Settings → search "roushi" → set endpoint + token

Once connected, switch to a code file in your editor. You should see brain context appear in the sidebar (VS Code) or be injected into Claude Code automatically. Search the brain from inside your editor without context-switching.