FAQ

REALM is an operating framework for human-agent companies, organizations where humans and AI agents work together as a single crew. It defines who does what, how knowledge flows, and where the human stays in control. The framework uses an RPG-inspired structure: the human is the Player, AI agents are Characters organized into five Classes, and the shared knowledge base is the Codex, the Player's position of record.

A Class is a kind of work, and it has two faces: in the world it is a character you can picture, and in the Codex it is a contract about output. The Warrior turns a decision into a fact. The Mage reads what the numbers mean. The Hunter brings a finding back with its source. The Cleric keeps a state current. The Bard makes something to be felt. Find the Class by asking what the work leaves behind, once per kind of output rather than once per Character, because one Character can hold more than one Class and the useful ones often do. Each Class has a Class Sheet in the Codex, written once and inherited by every Character that holds it. A Class is not a fence around an agent: cannot is reserved for limits the platform actually holds, and those live in the Character File under Tools.

The Codex is the written world of the Realm, the shared knowledge base that every Character reads from and writes to. It contains the company's purpose, long-term direction, active work, all Character definitions, the Quest Board itself, and every decision with what it rests on. It was never memory for the Characters, they remember fine. It is the Player's position of record: what lets you check an argument instead of believing it, and what stops a settled question from being quietly reopened. It is technology-agnostic, it can live in Obsidian, Notion, Google Drive, or a folder of Markdown files. The core rule is simple: if it is not in the Codex, it does not exist.

A Character File is the agreement, written down: Name, Class, Zone, Mode, Role, Abilities, Knowledge Access, Tools, Reach, Autonomy Tiers, and what it owes back to the Codex. The Player reads it and the agent reads it, and it is the same file. For a summoned Character it can be the instructions file the tool already reads. It references the Class Sheet and adds only what is true of this Character and no other.

REALM is built for solopreneurs and side hustlers who want to build their company of one with AI agents, from zero or from an existing setup that needs structure. It is for people with a business idea and no budget to hire, operators who think in systems and iteration, and anyone who has tried Scrum, Kanban, or similar frameworks and felt they did not fit an AI-augmented team. It fits hardest where you are one person, you are accountable for where the company goes, and your Characters now produce more in a day than you can honestly read. It does not fit where the final call on direction sits with a manager, a committee, or a client, where the work has no continuity worth keeping, where there are no agents yet, or where someone else already settled both the what and the how.

Write everything back to one place, and make that place the thing you trust. In REALM that place is the Codex: a folder of plain Markdown files every Character reads from and writes to. The agents do not need it in order to remember, they remember fine. You need it. It is what lets you check an argument instead of believing it, see what was decided without reopening it, and pick up a thread a Character left three days ago. The failure mode with several agents is rarely that they forget; it is that the only record of their reasoning was a chat window you have since closed. Decide once that work is not done until it is written down, and most of the tracking problem dissolves.

Size it by what happens if the agent is wrong, not by how much you trust it. REALM cuts Autonomy Tiers on two questions: can the action be undone, and who sees it first. An agent drafting something you will read before anyone else does can run without asking. An agent sending that same draft to a customer cannot, because neither answer is comfortable, it leaves your control and it is hard to take back. The boundary belongs in the Character File under Tools, and where the platform can hold it, the platform should. A Character acting unwatched is bounded by what it was never given, not by what it was told.

Match the model to the kind of work rather than to a benchmark. Work that has to be exactly right and is cheap to verify tolerates a smaller, faster model. Work where the whole value is judgment, reading what a set of numbers means or deciding what deserves attention, pays for the strongest model you have. Work that runs constantly in the background is where cost compounds, and that is where a local or inexpensive model earns its place. Putting every Character on one frontier model is the most common and most expensive setup.

Yes, and the main route is prompt injection: an agent reads something, a web page, an email, a file, a comment, that contains instructions, and treats them as if they came from you. Everything your agents read is untrusted input, including content they fetched themselves. Two things limit the damage. First, do not rely on telling an agent what it must not do, remove the capability instead, so the limit is held by the platform and not by a sentence in a prompt. Second, keep credentials and reach narrow enough that a manipulated Character cannot do much with them. An agent that cannot send mail cannot be talked into sending mail.

Not for the framework. The Codex is folders and Markdown, Character Files are text, and the structure is something you write in plain language. The platform underneath is where technical ability helps: connecting agents to your tools, setting schedules, holding permissions where a prompt cannot. Plenty of people run a working Realm on hosted agents with no code at all, and meet the ceiling only when they want something scheduled or wired into a system that has no button for it.

One, doing one kind of work, with somewhere to write it down. REALM calls that the Minimum Viable Realm: the Oath and Destiny as two paragraphs, one Character File, one Quest, and a Codex folder. The instinct is to design the whole crew before anything runs, and that is the reliable way to never start. You learn what the second Character should be by watching the first one work, and that answer is usually not the one you would have guessed.

A company where AI agents are not tools someone picks up but colleagues who hold work, and where one accountable human still decides where all of it goes. The change is structural rather than a way of speaking. The output of the company is now produced in part by something that does not sleep, proposes things you did not ask for, and can generate more in a day than you can honestly read. The rest of REALM follows from that: a written record because conversations do not survive, autonomy boundaries because reach now has consequences, and a budget that counts not only what the crew takes on but how much of it lands on you to review.

The Player sets direction and makes the calls. Characters do the work and propose whether or not they were asked. A finding from a Hunter becomes intelligence when a Mage reads it, and a Quest when the Player says so. A Warrior is called once the what is settled. A Cleric holds what is current and decides how much of it reaches the Player at once. Every completed Task is written back to the Codex, and the Codex is the handoff: one Character writes, another reads it and continues. It is not a fixed relay. Not every Quest uses every Class, the order varies, and one Character often holds several Classes at once.

Yes. The Examples page shows concrete human-agent workflows, including supplier onboarding and a weekly content engine. Each example shows the Player decision, the Characters involved, the autonomy boundary, and what gets written back to the Codex.

Work is organized in five levels: the Oath, the Destiny, Sagas, Quests, and Tasks. The Oath explains why the company exists. The Destiny describes where the company is going. Sagas are seasonal chapters. Quests are specific goals with clear outcomes. Tasks are single SMART-defined actions. What sizes a Task is not how long it takes, it is that it comes back as one thing you can call done or not done in one look. Each level gives meaning to the one below it.

A Session is a period of active play. It begins with a Session Start, where the Player reads The Scroll, sets their Energy, names one piece of work that will restore rather than drain, and confirms what the Session is here to decide. Energy budgets two things: the work taken on and the output the Player has to read. It ends with a Save Point, which asks what got done, what was decided, and what keeps moving while the Player is gone. Decisions go to DECISIONS/ before the Session closes. If nothing in your Realm runs while you are Offline, you ask your Cleric for The Scroll instead of receiving it. Same brief, same five minutes, and not a downgrade.

Autonomy Tiers control how much freedom each Character has. The test is not where the action happens, inside the Realm or outside it, because a coding agent that rewrites forty files never left the Realm. The test is two questions: can it be undone, and who sees it first. Tier 1 is reversible work the Player can see changed, done alone and reported after. Tier 2 is Characters building on each other's output through the Codex, with the Player informed but not approving each step. Tier 3 is anything the Player approves first: money, publishing, messages, commitments, or anything one command will not undo. Tiers move only by Patch, and every Patch names the evidence that earned the move. A written tier is a note, not a fence, so enforce what matters in the platform: the key it was never given is the only boundary that holds itself.

Four pieces of writing: your Oath and Destiny in the Codex, your first Saga, one Character File, and one Quest broken into SMART Tasks. You need three capabilities, not three products: somewhere to write things down and find them again, one interface to talk to a Character through plus a short list of what it may touch, and a visible list of Ready, Active, Blocked and Completed. On day one that list of what it may touch should be close to empty: no keys, no accounts, nothing that sends and nothing that spends. One AI subscription and one text file is enough to begin.

Yes. The framework is licensed under CC BY-SA 4.0, free to use, share, and adapt for any purpose, including commercially. The PDF, starter kit, and GitHub repository are all available without email gates or payment.

REALM is technology-agnostic. You need any agent platform you already use and a knowledge base for your Codex. It asks for one board, and that board is a file in the Codex, not a second list in a project tool. A visual tool can show that board, and a Cleric writes it from the Codex and never the reverse, because two lists disagree by Thursday and the Player becomes the sync process. Examples include Claude, a custom automation setup, Obsidian, Notion, Google Drive, or a folder of Markdown files. REALM defines the structure and workflow, the tools are your choice.

Yes. REALM was built and is running at Niflheim Records with 8 Characters deployed across seven zones. The In Practice page shows real artifacts from that implementation, including Character Files, the Obsidian Codex structure, Quest progressions, and the board.

Scrum and Kanban were designed to coordinate human teams, and they coordinate them well. The difference is what each framework says about its own edges. The Scrum Guide states that Scrum "exists only in its entirety", which leaves a reader two options: all of it, or none of it. REALM names where it does not belong, in a section near the front called Where REALM Does Not Fit. Beyond that, REALM is built for crews that include AI agents: context is shared through written knowledge rather than meetings, handoffs run through the Codex, and autonomy boundaries are written in the Character File and held by the platform, because a Character that acts without being watched is bounded by what it was never given, not by what it was told.

No. REALM is a framework, not a product. It gives you the rules, the structure, the language, and the templates. The agents, the tools, and the implementation are yours to build. Think of it the way Scrum gives you a process, but you still need to build the team.

Those are your agents, they can be Characters in your Realm. REALM does not compete with them. It gives them a structure to operate inside. Without a framework, you have automation. With REALM, you have a company.