Arche

An AI-native system where your knowledge, tools, processes, and expert agents work together.

Arche is an open source AI agent platform for companies or individuals who want to build a shared ecosystem of agents, knowledge, and workflows for their team, or use the same system as a personal second brain.

• 🏢 For companies: build a shared AI-native ecosystem of agents, knowledge, and processes that every member of your team can use.
• 🧠 For personal use: run Arche solo as a local AI-native second brain for notes, docs, and ideas.

From one place, you can run support, copywriting, SEO, marketing, research, requirements, and ops agents. Instead of starting every prompt from zero, you define your products, tone, processes, docs, and source material once. Each agent gets that shared knowledge plus its own isolated workspace, so your AI team stays consistent and useful.

Arche also works with the tools you already use. Built-in connectors cover Google Workspace, Linear, Notion, Zendesk, Ahrefs, Umami, and Meta Ads; Slack is available as an admin-managed integration; and compatible remote MCP servers let you extend the stack further. Arche becomes a shared entry point into your digital stack, not just another AI editor.

See Arche

Arche usually fits one of these patterns:

What Arche Is For

• 🏢 Teams: run a shared crew of experts for support, copywriting, SEO, marketing, research, requirements, and ops.
• 🔌 Connected stack: plug Arche into Google Workspace, Linear, Notion, Zendesk, Ahrefs, Umami, Meta Ads, Slack, and compatible remote MCP servers.
• 🧠 Individuals: use the same system as an AI-native second brain for notes, docs, and ideas.
• 💻 Deployment choice: start on one machine with Desktop or self-host Arche for your whole team.

How Arche Works

1. Choose a template and the agents you want to run.
2. Add your company or personal knowledge, tone, and working context once.
3. Connect the tools you already use.
4. Let each agent work in its own isolated workspace without losing the shared context.

Principles

• 🤖 AI-native: your knowledge base is meant to be usable by AI from day one, not bolted on later.
• 🏢 Team-first: Arche is designed around shared context, reusable experts, and repeatable work across a company.
• 🔌 Connected stack: built-in integrations and compatible remote MCP servers let Arche work with the tools you already use.
• 🧩 Specialized agents: support, copywriting, SEO, marketing, research, requirements, and ops can share knowledge without sharing runtime state.
• 🧠 Individual-friendly: the same model also works as a personal second brain on one machine.

Start Here

I want to...	Best option
🏢 Run Arche for a team	Self-Hosting
🧠 Use Arche locally or build a personal second brain	Desktop App
🛠️ Work on the codebase	Local Development

Desktop App

If you want the fastest path, start with the desktop app. It runs on your machine with no server or Docker setup required.

Desktop vault behavior:

• Arche Desktop opens directly into the workspace for the last valid vault.
• If no vault is selected yet, it opens a launcher where you can create or open a vault.
• Each vault is a visible folder that contains its own database, KB repos, runtime state, and secrets.
• Opening another vault launches a separate Electron process and window, similar to Obsidian.

Breaking change:

• Desktop no longer reads or migrates legacy hidden data from ~/.arche or ~/.arche-opencode.
• You must create or open a visible Arche vault folder.

Download

Head to the latest release and download the installer for your platform:

Platform	File
macOS (Apple Silicon)	Arche-arm64.dmg or Arche-<version>-arm64-mac.zip
macOS (Intel)	Arche-x64.dmg or Arche-<version>-mac.zip

Official GitHub release assets are currently published for macOS only.

Linux and Windows packaging targets exist in apps/desktop, but they are not part of the current release workflow or validation matrix, so they should not be documented as supported release artifacts.

Build from Source

If you prefer to build the desktop app yourself:

# Prerequisites: Node.js 24+, pnpm 10+, Go 1.22+

# 1. Install dependencies
cd apps/web && pnpm install
cd ../desktop && pnpm install

# 2. Build a desktop package for your current host platform
cd ../..
bash scripts/build-desktop.sh

The packaged desktop artifacts will be in apps/desktop/release/.

For more details, see apps/desktop/README.md.

Self-Hosting

If you want Arche for a team, self-host it on your own infrastructure.

One-Click DigitalOcean Install

For the narrow-path setup, there is now a one-click installer that creates a fresh DigitalOcean Droplet, configures Docker, deploys the latest Arche images, auto-generates secrets, and exposes the app on a nip.io hostname.

curl -fsSL https://thearcheproject.com/install | bash

You can also pass inputs up front:

curl -fsSL https://thearcheproject.com/install | bash -s -- --token "$DIGITALOCEAN_TOKEN" --email admin@example.com --version v2.0.0

The installer prompts for:

• DigitalOcean API token
• Email address for the initial admin account and Let's Encrypt

You do not provide server, database, or admin passwords. The Go deployer generates a local SSH keypair for the deployment, the Droplet generates the runtime secrets during bootstrap, and archectl fetches the recovery details back over pinned SSH into ~/.arche/deployments/.

The shim installs archectl into /usr/local/bin when that directory is writable, otherwise into ~/.local/bin. After installation, use the same binary for lifecycle commands:

archectl install --token "$DIGITALOCEAN_TOKEN" --email admin@example.com --version v2.0.0
archectl update --token "$DIGITALOCEAN_TOKEN" --version v2.0.0
archectl destroy --token "$DIGITALOCEAN_TOKEN"

By default, archectl keeps output minimal and shows only lifecycle steps plus in-place progress. Add -vv or --verbose to show SSH/bootstrap logs.

GitHub KB Sync

Admins can connect the deployment-wide shared knowledge base to a GitHub repository from Settings -> Integrations -> GitHub KB sync.

• Arche creates a GitHub App from a manifest, so admins do not need to manually copy app IDs or private keys.
• The GitHub App needs Contents: Read and write and Metadata: Read-only permissions for the selected repository.
• Sync scope is only KB content. Runtime state, databases, provider credentials, connector secrets, and user sessions are not pushed to GitHub.
• Starting with an empty repository is recommended. The first sync pushes the local KB into the selected GitHub branch.
• Publishing from the Knowledge Review panel fetches the latest GitHub state before pushing. Clean merges are applied automatically.
• Merge conflicts are surfaced in the existing Knowledge Review panel and resolved with the same local/incoming/manual flow used for KB conflicts.

If the local state file is missing, recovery flags are available:

archectl update --token "$DIGITALOCEAN_TOKEN" --version v2.0.0 --ip 203.0.113.10 --ssh-key ~/.arche/deployments/arche-20260410-120000-ssh.pem
archectl destroy --token "$DIGITALOCEAN_TOKEN" --droplet-id 123456789 --firewall-id firewall-id --yes

Legacy password-based recovery remains available for older deployments with --ssh-password.

Assumptions:

• DigitalOcean only
• The shell entrypoint downloads https://github.com/peaberry-studio/arche/releases/latest/download/archectl_<os>_<arch> for macOS/Linux on amd64/arm64
• Image tags are derived from --version: latest by default, or a pinned tag such as v2.0.0
• Public URL is derived automatically as https://arche-<droplet-ip>.nip.io
• Local deployment state is stored at ~/.arche/deployments/current.json

Operational caveats:

• The default public URL depends on the third-party nip.io DNS service. If nip.io is unavailable, point your own DNS name at the Droplet before relying on the deployment.
• archectl destroy permanently removes the Droplet and attached data volumes. Create a Droplet snapshot or other backup before destroying a deployment you may need to recover.

To test installer changes locally before publishing a GitHub release, build the matching binary into /tmp, point the shim at that directory, and run template validation:

cd infra/one-click
GOOS="$(go env GOOS)" GOARCH="$(go env GOARCH)" go build -o "/tmp/archectl_$(go env GOOS)_$(go env GOARCH)" .
cd ../..
ARCHECTL_RELEASE_BASE_URL=file:///tmp bash install.sh --validate-only

Deploy to a VPS (Ansible)

One-command deployment with automatic TLS, database provisioning, and secrets management.

See the full guide: infra/deploy/README.md

Deploy with Coolify

If you use Coolify to manage your infrastructure, Arche has first-class support with zero-downtime rolling updates.

See the full guide: infra/coolify/README.md

Local Development

If you want to work on Arche itself, set up the full development stack locally with hot reload:

# Prerequisites: Node.js 24+, pnpm 10+, Podman (or Docker) with Compose

cd infra/deploy
cp .env.example .env
./deploy.sh --local-dev

Open http://arche.lvh.me:8080 and log in with admin@example.com / change-me.

For step-by-step instructions, see infra/compose/README.md.

Documentation

Document	Description
ARCHITECTURE.md	Technical architecture, tech stack, data model, and source code structure
CONTRIBUTING.md	Contribution guidelines
apps/web/README.md	Web application setup and internals
apps/desktop/README.md	Desktop app development and packaging
infra/deploy/README.md	VPS deployment guide (Ansible)
infra/coolify/README.md	Coolify deployment guide
infra/compose/README.md	Local Podman Compose stack
infra/workspace-image/README.md	Workspace container image

Test Coverage

Current line coverage is tracked in apps/web with Vitest.

• Overall coverage: cd apps/web && pnpm coverage
• Unit coverage: cd apps/web && pnpm coverage:unit
• Integration coverage: cd apps/web && pnpm coverage:integration
• Refresh README badges: cd apps/web && pnpm coverage:refresh
• Badge verification in CI: automatic on pull requests targeting main

Browser E2E tests in apps/web/e2e and apps/desktop/e2e still run with Playwright, but they do not publish a reliable line-coverage percentage yet. There are also environment-dependent backend E2E tests in apps/web/src/**/*.e2e.test.ts and smoke tests in apps/desktop.

License

This project is licensed under the Apache License 2.0. See LICENSE.