Arcana lets you run AI agents as tarot card archetypes. Pick a card and your agent has a personality, a focus, and a tone that stay consistent, every session. No prompt engineering required.
Free forever. Apache 2.0 licensed. macOS & Linux.
They drift without a consistent character, you have no overview of what they're doing, and no one is in charge. You can't tell what they know, why they behaved the way they did, or who's responsible when one goes off-track.
Out of the box, an agent is just a model and a wall of instructions, with no archetype and nothing that makes one reliably different from the next. So the same tool hands you a different answer every time, and nothing explains why.
Personality evaporates between prompts. The same agent runs differently every session because nothing anchors who it is, just instructions you rewrite by hand.
Nothing records why, so you see the output but not the reasoning, the cost but not the cause. When something breaks, you're debugging a black box.
Arcana is built on one premise:
agents should have archetypes, not just prompts.
Each of the 22 Major Arcana is a complete behavioral archetype: enough structure to make your agent reliably itself, session after session.
One card sets the archetype. Add modifier cards and Arcana blends them into one personality, temperature and tone adjusted, and flags any tension before it ever runs.
The primary card defines the agent's full archetype. Modifier cards shift the temperature at the weights shown and add a targeted tone layer. It all blends into a single personality once, instead of being patched together each time the agent runs.
Arcana reads your combination like a spread. Archetypes that amplify each other get a synergy flag, and conflicting ones get a warning before you run it.
The modifier recomputes temperature, not just tone. Hermit (0.35) with a touch of Sun (0.75) runs measured but brighter. You can see exactly how it was set, it comes out the same every time, and you can override it whenever you want.
Local or hosted, your choice. Your keys never leave your machine.
Browse the 22 Major Arcana. Each carries a defined archetype, temperature, and tone.
Name it, pick its card, and Arcana sets everything up: its personality, temperature, and tone.
From the reckless Fool to the all-seeing World: not a list of features. A cast of characters.
Common starting points: Hermit (IX): research agent · Empress (III): writing agent · Magician (I): automation agent · Justice (XI): code review
Your agents remember: what you told them, what they worked out, what the whole system has learned, kept across every session. You decide what stays with one agent and what the group shares. Arcana owns its memory; you own your knowledge.
The agent just remembers. It never has to know where things live.
Everything one agent has picked up: what it did, what it learned, what you asked it to remember. Nobody else sees it.
A common notebook a group of agents reads and adds to. Perfect for a set of agents working the same problem together.
What the whole system should know. Every agent can read it, and the most important notes land here on their own.
As your agents work, the useful bits get saved automatically, with no database to set up and nothing to wire together by hand.
Ask a question and it surfaces the right memories, gathered from everywhere the agent knows, most relevant first.
An agent's own memory is protected above all else. If anything else is briefly unavailable, it keeps going instead of breaking.
It matches what you mean, not only what you typed. Ask about a 10K and it connects that to the note where you said you like metric units.
Under the hood: local per-agent database, optional AI-powered search, nothing leaves your machine.
Point Arcana at a folder of notes, like an Obsidian vault, and your agents can read it. It only reads, it never changes your files. Links between your notes become connections the agents can follow.
Ollama, Anthropic, or any OpenAI-compatible endpoint. Local models work out of the box.
A tool and MCP layer, so a card's tool preferences become real capabilities agents can reach for. Shipping soon.
A meta-agent that routes tasks, audits for reversed cards (agents running outside their archetype), and writes your daily briefing. The one card that cannot be reversed. Shipping soon.
Start local in one command. The deck is waiting.