Aurite AI

---

The Problem

Every time you start a new conversation with your AI copilot, it forgets everything.

• 🔄 You repeat the same context about your project, your team, your standards
• 🤷 The copilot makes mistakes you've already corrected in past sessions
• 📄 Your policies, specs, and business rules sit in files the copilot never sees
• 🧠 Decisions and rationale from past conversations are lost forever

Copilots are powerful — but they have amnesia.

The Solution

Kai gives your copilot a persistent memory that grows smarter over time.

Without Kai	With Kai
Copilot starts fresh every session	Copilot remembers what it learned
You repeat context manually	Context surfaces automatically
Knowledge lives in your head	Knowledge lives in a structured KB
Decisions are forgotten	Decisions persist across sessions

How it works: Kai runs as an MCP server alongside your copilot. You teach it your context once — policies, specs, decisions, patterns — and it proactively surfaces relevant information when you need it.

🔒 All data stays local. Your code and context never leave your machine.

---

Quickstart (Claude Code)

Step 1: Add Kai to Claude Code

claude mcp add kai -s user -e ANTHROPIC_API_KEY="your-anthropic-api-key" -- npx @aurite-ai/kai

Scope options:

• -s project — Config stored for current project only
• -s user — Config stored globally (available across all projects)

Step 2: In any project, tell your copilot:

"Set up Kai"

This deploys copilot rules and runs onboarding. The copilot asks a few questions to understand your context — this only happens once.

Step 3: Start teaching it your context:

"learn ~/Downloads/api-guidelines.pdf"

"learn the docs/ folder"

Step 4: Start working — Kai surfaces the right context automatically.

"build a customer support agent"

Kai feeds your copilot your API conventions, auth patterns, and related context. No reminders needed.

npm (Global Install)

npm install -g @aurite-ai/kai

Configure your MCP client to use kai-mcp as the command.

npx (No Install)

npx @aurite-ai/kai

Docker

docker pull kai/mcp
docker run -i kai/mcp

From Source

git clone https://github.com/Aurite-ai/kai.git
cd kai
pnpm install
pnpm --filter @aurite-ai/kai build
pnpm --filter @aurite-ai/kai bundle

---

What It Looks Like

You teach Kai your company's context:

"learn ~/docs/api-guidelines.pdf"

"learn the docs/ folder"

Later, you start a task:

"build a customer support agent"

Kai automatically surfaces the relevant context to your copilot:

• ✅ Your API conventions and auth patterns
• ✅ Customer data models and access policies
• ✅ Error handling and response format standards
• ✅ Related endpoints already in the codebase

Your copilot builds it right the first time — no reminders needed.

---

How It Works

┌──────────────────────────────────────────────────────────────────┐
│  YOU                          COPILOT                  KAI      │
│                                                                  │
│  "set up Kai"     ──────────────────►  deploys rules  ─────►  .kai/    │
│                               asks questions          stores    │
│                                                       context   │
│                                                                  │
│  "learn these docs" ────────────────►  kai_learn      ─────►  knowledge │
│                                                       base      │
│                                                                  │
│  "build feature X" ─────────────────►  kai_prepare    ─────►  surfaces  │
│                               _context                relevant  │
│                                                       files     │
└──────────────────────────────────────────────────────────────────┘

---

💡 If Kai saves you from repeating yourself, consider giving it a ⭐. It helps others discover the project.

---

Contents

• The Problem
• The Solution
• Quickstart
• What It Looks Like
• How It Works
• How It Compares
• Features
• Available Tools
• Documentation
• Migrating from Kahuna
• Contributing
• License

---

How It Compares

Feature	Kai	Copilot Memory	RAG Tools	Manual Context
Persists across sessions	✅	Partial	✅	❌
Learns from files & conversations	✅	❌	Files only	N/A
Proactive context surfacing	✅	❌	Query-based	❌
Auto-classifies knowledge	✅	❌	❌	Manual
Works across projects	✅	❌	Varies	❌
Zero-config for copilot	✅	✅	❌	❌
Data stays local	✅	❌	Varies	✅

Kai is not a replacement for built-in copilot memory — it's what copilot memory should have been.

---

Features

• 🧠 Knowledge Base — Store, categorize, and retrieve context from markdown files
• 🎯 Smart Context Surfacing — Automatically surface relevant knowledge for your task
• 🔗 Integration Management — Discover, verify, and use external service integrations
• 🔐 Secure Credential Vault — Store and manage secrets with multiple provider support
• 📊 Usage Tracking — Monitor token consumption and costs per project
• 🚀 Onboarding System — Guided setup for organization and project context

Available Tools

Tool	Description
kai_initialize	Deploys copilot rules, runs onboarding
kai_learn	Adds files to knowledge base with classification
kai_prepare_context	Surfaces relevant knowledge for a task
kai_ask	Quick Q&A against the knowledge base
kai_delete	Remove outdated files from the knowledge base
kai_provide_context	Store org or user context in the knowledge base
kai_usage	View token usage and cost summary for the project
kai_list_integrations	List all discovered integrations and their status
kai_use_integration	Execute operations on discovered integrations
kai_verify_integration	Verify integration credentials and connectivity
kai_discover_integration	Discover and add a connector for a new service, API, or tool
kai_update_connector	Update or refresh an existing integration connector
health_check	Verify MCP server connectivity

---

Documentation

For Users:

• MCP Server Documentation — Installation, tools, configuration
• Advanced Documentation — Integrations, vault, KB structure

For Contributors:

• Product Design — Core concepts, tool specifications
• Architecture: Repository Infrastructure
• Architecture: Context Management System

---

Migrating from Kahuna

Kai was previously named Kahuna. Kai does not read any of the old Kahuna paths or settings, so everyone with an existing install needs to do these steps once.

Before you start: check that ~/.kahuna holds Kai data (knowledge/, integrations/, connectors/, .env) and not files from another Kahuna project.

1. Update your local clone (contributors only)

git remote set-url origin https://github.com/Aurite-ai/kai.git
git checkout main && git pull
pnpm install
pnpm clean && pnpm build

2. Move your data directory

mv ~/.kahuna ~/.kai

3. Rename environment variables and secrets

Every KAHUNA_* variable is now KAI_* (for example, KAHUNA_KNOWLEDGE_DIR is now KAI_KNOWLEDGE_DIR).

# Secrets stored by Kai
sed -i '' 's/KAHUNA_/KAI_/g' ~/.kai/.env

# Local MCP server config (contributors only)
sed -i '' 's/KAHUNA_/KAI_/g; s/\.kahuna-knowledge/.kai-knowledge/g' apps/mcp/.env

# Find leftovers in your shell profile, then rename them by hand
grep -n KAHUNA_ ~/.zshrc ~/.bashrc ~/.profile 2>/dev/null

On Linux, use sed -i instead of sed -i ''.

If you used apps/mcp/scripts/setup-claude.sh, also rename the repo-local knowledge base: mv .kahuna-knowledge .kai-knowledge.

4. Re-register the MCP server

Remove the old server:

claude mcp remove kahuna -s user
claude mcp remove kahuna -s project

Then add Kai back, using the published package or your local build:

# Published package
claude mcp add kai -s user -e ANTHROPIC_API_KEY="your-anthropic-api-key" -- npx @aurite-ai/kai

# Local build of this repo (contributors)
pnpm mcp:setup

Restart Claude Code, run claude mcp list, and confirm kai is connected and kahuna is gone.

5. Update projects that use Kai

Tool names changed from kahuna_* to kai_* (for example, kahuna_learn is now kai_learn). In each project where Kai was set up:

[ -d .kahuna ] && mv .kahuna .kai                                # usage history and context guide
[ -f .kahuna-test.json ] && mv .kahuna-test.json .kai-test.json  # test projects only

Then refresh the copilot rules so they use the new tool names. kai_initialize skips files that already exist, so run it with overwrite:

1. Commit or back up any rules you customized (for example, .claude/CLAUDE.md).
2. Tell your copilot: "Run kai_initialize with overwrite=true"
3. Review the changes with git diff and restore any customizations.

Also replace any .kahuna entries in the project's .gitignore with .kai.

6. Verify

ls -d ~/.kahuna 2>/dev/null
grep -n KAHUNA_ ~/.kai/.env apps/mcp/.env 2>/dev/null

Neither command should print anything.

---

Contributing

We welcome contributions of all kinds!

• 🐛 Found a bug? Open an issue
• 💡 Have an idea? Open a feature request
• 🔧 Want to contribute code? Open a PR

Prerequisites

• Node.js 18+
• pnpm 9+

Quick Start

# Install dependencies
pnpm install

# Set up environment
cp apps/mcp/.env.example apps/mcp/.env

# Build workspace packages
pnpm build

# Run tests
pnpm test

Scripts

Command	Description
pnpm build	Build all packages (via Turborepo)
pnpm test	Run all tests across workspace
pnpm lint	Lint codebase (Biome)
pnpm lint:fix	Lint and auto-fix issues
pnpm format	Format codebase (Biome)
pnpm typecheck	Type-check all packages
pnpm clean	Remove build artifacts and caches

Testing CLI

Command	Description
pnpm kai-test	Run testing CLI
pnpm test:create	Create a test project from a scenario
pnpm test:list	List available scenarios and test projects
pnpm test:collect	Collect results from a test session

Project Structure

kai/
├── apps/
│   └── mcp/                # MCP server (stdio) — context management tools
│       ├── src/
│       │   ├── knowledge/  # Knowledge base domain logic (agents, storage, surfacing)
│       │   ├── integrations/   # External service integration management
│       │   ├── vault/      # Secure credential management
│       │   ├── usage/      # Token usage and cost tracking
│       │   └── tools/      # MCP tool handlers
│       └── templates/      # Project initialization templates
├── packages/
│   ├── testing/            # QA testing infrastructure (scenarios + CLI)
│   └── vck-templates/      # Copilot configuration templates
└── docs/                   # Documentation

---

License

MIT