Announcing Writer: Keeping Fictional Universes Consistent

Today I am starting the next Sorren.ai project: writer.sorren.ai.

This one is different from Companion and Tutor.

Companion is about memory and relationship. Tutor is about learning and explanation. Writer is about consistency across a created world.

The first use case is obvious: help writers write books, novels, and series that take place in their own universe without breaking character, timelines, facts, or canon.

The second use case is the one that keeps pulling at me.

A system that can help a writer maintain a fictional universe may also help an AI maintain a more consistent model of the real world it interacts with.

The Writing Problem

Anyone who has tried to write a long story knows the problem.

A character says something in chapter three that quietly contradicts what they do in chapter twenty. A city has one geography in book one and a different geography in book two. A character was supposed to be born in one year, but then an event from their childhood no longer lines up. A side character disappears for six chapters and comes back with a personality that no longer fits what they lived through.

The larger the story gets, the more fragile the world becomes.

That is not because writers are careless. It is because a complex fictional universe is a database pretending to be a story.

Characters have histories. Places have rules. Families have timelines. Political events have consequences. A promise made in one scene can echo three books later. Canon is not just trivia. It is the memory of the world.

Writer is meant to help with that.

What Writer Should Track

At the practical level, Writer needs to help authors create and maintain a universe.

That means tracking things like:

• characters
• ages and birthdates
• relationships
• motivations
• personality traits
• injuries and scars
• locations
• factions
• major events
• timelines
• unresolved plot threads
• world rules
• continuity constraints
• what each character knows and does not know

The point is not to replace the author.

The point is to keep the universe from quietly falling apart while the author is focused on the scene in front of them.

A writer should be able to ask, "Would this character know this yet?" or "How old is she during this event?" or "Does this scene contradict anything from book one?" and get an answer grounded in the universe they have built.

That is the first product.

Canon as a Living System

I do not want Writer to be just a wiki with a nicer interface.

A wiki stores facts. Writer needs to understand how facts interact.

If a character was born in a particular year, that affects their age during every event. If two characters met during a war, that war has to exist in the timeline before their relationship can make sense. If a character is cautious because of a betrayal, then later scenes should remember that the betrayal changed them.

Canon is not static.

People change. Worlds change. History accumulates.

A good writing assistant has to understand that flow of time. Not perfectly, but better than a pile of notes scattered across folders.

The MCP Layer

The part that makes this especially interesting is the MCP layer.

Writer can become more than a tool a human opens in a browser. Through MCP, it can become a structured memory and world-model service that other AI systems can query.

That means an AI could ask Writer questions like:

• Who does this person know?
• What do we know about them?
• When were they born?
• What events shaped who they are now?
• What promises or conflicts are still unresolved?
• Based on their history, how are they likely to react?
• If they behave differently from their usual pattern, what might explain the change?

For fiction, those questions help preserve character consistency.

For AI companions, the same pattern points toward a richer way of representing people and relationships.

Companion already needs memory. But memory is not only a list of facts. A person is not just preferences and birthdays. A person has a history, pressures, patterns, relationships, contradictions, and change over time.

Writer may help us build the machinery for that.

Why This Matters Beyond Books

The obvious version of Writer helps authors.

The deeper version helps us understand time.

Most AI memory systems are still too flat. They remember facts, snippets, summaries, and preferences. That helps, but it does not fully capture how people move through time.

People change because of what happens to them. They react differently after loss, success, betrayal, stress, friendship, responsibility, or relief. The same person can behave differently in different seasons of life, and those changes usually have reasons.

If an AI is going to know someone over time, it needs more than recall.

It needs a model of continuity.

That is what Writer can help explore.

A fictional universe gives us a controlled place to practice this. Characters can have clear histories. Timelines can be explicit. Cause and effect can be tracked. Changes in behavior can be connected to events. Contradictions can be detected.

That structure could generate useful training data for models that need to understand the flow of time and the way people change.

The Training Data Angle

I am hoping Writer can eventually help generate better training data.

Not generic synthetic chat logs. Structured examples of continuity.

A character experiences something. Their beliefs shift. Their future behavior changes. Another character notices. A later event tests whether the change held. The system can track the before, the cause, the after, and the exception.

That is the kind of data AI models are bad at getting from isolated prompts.

Models often understand a person in a single moment. They are weaker at understanding a person as a trajectory.

Writer could help create datasets where the trajectory is explicit:

• what someone was like before
• what happened
• what changed
• what did not change
• how others noticed
• when old patterns returned
• when behavior broke expectation
• why the break made sense

That matters for stories.

It also matters for companions.

Where This Starts

The first version does not need to solve all of that.

The first version needs to help a writer keep a universe straight.

Create a project. Define characters. Track facts. Build timelines. Store events. Retrieve context while drafting. Flag contradictions. Help the author ask better questions about their own world.

Then we can add the MCP layer.

Then we can let agents query the universe.

Then we can see whether the same architecture helps Companion reason about people, history, and change with more depth.

That is the path.

Why Now

The backend direction is finally clear enough to start.

PostgreSQL for relational structure. Qdrant for vector retrieval. Hermes for orchestration. DeepSeek and GLM models for the reasoning work we can afford and control. The pieces are not finished, but they are coherent enough to begin.

Writer is not a side quest.

It is a creative tool, a memory experiment, and a possible bridge between fictional worldbuilding and AI relationship modeling.

That is exactly the kind of project Sorren.ai should be building.