A local-first AI agent desktop app

Arcana runs locally on your machine with a Web UI and a tools system. You can connect different model providers while keeping API keys encrypted and stored locally, turning your computer into a controllable AI agent workspace.

Local Web UI without relying on hosted chat services.
Supports multiple model providers, configurable in the UI or config.
API keys are stored in an encrypted local vault, not in code or logs.
Open-source project that you can inspect and customize locally.

Download macOS installer View install and checksum steps

The primary experience today is on macOS; you can also use the CLI and Web UI commands from this repo on other platforms where Node.js is available.

Goal → Plan → Execute

Set a goal

Describe your final goal in natural language, such as "compile a technical research brief" or "design an automated workflow for this project".
Generate a plan

Arcana breaks the goal into a multi-step plan, including which tools to call, which sites to visit, and what intermediate outputs to expect, so the process stays visible and editable.
Execute with tools

After you review the plan, Arcana calls browsers, scripts, and APIs step by step. You can pause, modify, or take over any step at any time.

How Arcana works: from goal to execution

Arcana turns natural-language requests into repeatable agent task flows: align on the goal, generate a structured plan, then execute it with tools and models.

1. Clarify the goal

Start from a complex task or a common pattern such as research, coding assistance, or data cleanup. Arcana helps clarify boundaries, constraints, and priorities before executing.

2. Create a structured plan

The agent decomposes the work into steps, annotating each with tools, external resources, and expected intermediate artifacts. You can edit or extend the plan before running it.

3. Execute with tools

Arcana executes the plan using its tool system: browse the web, run scripts, call HTTP APIs, or read local files. Every step is logged in the UI for inspection and troubleshooting.

Capabilities for real-world work

Arcana is more than a chat box. It is composed of capabilities you can mix and match depending on how you prefer to work.

Multiple models and providers

Configure different models and providers such as OpenAI via config files or the UI, and switch between reasoning, generation, and embedding models per task.

Secure key management

Manage credentials such as providers/openai/api_key through a Secrets UI. Arcana stores them encrypted in an internal vault instead of committing them to git or writing them to logs.

Tooling and plugins

Use the tools system to let agents browse the web, run scripts, call HTTP APIs, and more. Extend it with custom tools so your existing services can be reused safely inside agent workflows.

Memory and context

Arcana can maintain long-term memory for the default agent (for example in a ~/.arcana/agents/default/MEMORY.md file), helping it remember important facts and preferences across sessions.

Local Web UI

Interact with agents through a local Web UI instead of a hosted chat service. Start the Web UI from the command line and connect to a local port in your browser.

Automation and observability

The UI shows each step of an agent run: which tools were used, which URLs were requested, and what intermediate artifacts were created, so you can audit and refine behavior.

Typical use cases

Whether you are an engineer, a product manager, or an individual creator, you can treat Arcana as an "AI collaborator that can act". Here are some representative patterns.

Code exploration and refactoring

Ask the agent to read repositories, explain module relationships, and draft code or refactoring suggestions, while you keep full control over reviews and merges.

Technical and market research

Use the browser tools to collect information from official docs, blogs, and community discussions, then have the agent structure it into summaries, comparison tables, and next-step suggestions.

Information cleanup and knowledge bases

Collect scattered notes, issues, and docs into Arcana, and let the agent clean, classify, and convert them into draft knowledge base content for you to review and publish.

Personal productivity and scripts

Start from repetitive daily tasks and let the agent propose and run scripts or workflows—for example, organizing downloads, renaming files in bulk, or generating recurring reports.

Download, verify, and install Arcana

Arcana offers a .dmg installer for macOS. The section below outlines the download link and recommended checksum verification steps so you can confirm the integrity of the file.

1. Download the macOS installer

Use the main button below to download the Arcana-latest.dmg installer from the /downloads/ directory. In this repository it is a placeholder path; for a production site you should upload a real installer file there.

Download Arcana-latest.dmg

Note: the repository ships with Arcana-latest.dmg.placeholder.txt as a placeholder. When publishing, replace it with a real .dmg and update the checksum file.

2. Verify the SHA-256 checksum

After downloading in your browser, make sure the file lives in your download directory (for example ~/Downloads).
Open a terminal and change into the download directory:
```
cd ~/Downloads
shasum -a 256 Arcana-latest.dmg
```
Open site/downloads/Arcana-latest.sha256 from this repository and compare the hash printed in your terminal with the one recorded there.
Only continue with installation if the hashes match. If they do not, download again or confirm the installer source before proceeding.

3. Install and launch

Double-click Arcana-latest.dmg.
Drag the Arcana app into your Applications folder.
Open Arcana via Launchpad or Spotlight.
If macOS warns about an unidentified developer, approve the app under "System Settings → Privacy & Security" and launch it again.

Checksum file location

The recommended SHA-256 checksum lives at site/downloads/Arcana-latest.sha256. When you publish a new installer, update this file with the correct hash.

The suggested format is one line containing the hash followed by the filename, for example xxxxxxxx... Arcana-latest.dmg. The value in this repository is a placeholder and must be replaced before release.

Privacy and security notes

Model provider keys are entered through Arcana's Secrets UI and stored in an internal encrypted vault.
By default, keys are not written into files in the repo and are not logged.
You can remove or rotate keys at any time and change how agents call external services according to your own security policies.

Frequently asked questions

A few common questions about Arcana, model configuration, and security are collected here so you can get oriented before diving in.

What is Arcana?

Arcana is a local AI agent system with both CLI and Web UI entry points. It focuses on goal → plan → execute workflows rather than just single-shot chat.

Is Arcana free?

The Arcana code in this repository can be used without an additional license fee. When you connect third-party model providers (such as OpenAI), you must still follow their pricing and terms.

Is Arcana open source?

Yes. Arcana is an open-source project, and you can review the code and adapt it locally as long as you follow the project license.

Are my API keys uploaded anywhere?

By design, keys entered via the Secrets UI are kept encrypted on your machine and are not sent to a hosted Arcana backend. You are still responsible for managing them according to your security practices.

Can I use Arcana offline?

The Web UI and agent logic can run locally, but if you use cloud-hosted models, network access is still required for model calls. You can also connect self-hosted or local models if you prefer.

How is this different from hosted chat products?

Hosted chat products are optimized for one-off conversations. Arcana is closer to a programmable local AI agent: tools, configuration, and memory live on your machine and can be reused as workflows.

How do I use Arcana on other platforms?

This page focuses on the desktop experience, but you can also use the CLI entry points from the repo, such as node ./bin/arcana.js chat and node ./bin/arcana.js web serve, subject to the documentation.