Autonomous organizations
Observe how different agent policies reshape the same company.
Run one policy that centralizes approvals and another that pushes decisions to teams, then read how hiring, spending and customer commitments drift apart.
Every action weavesanother world.
WorldBraid turns APIs, databases, tools, documentation and traces into executable worlds. Place an agent at the Origin and watch every decision create a different future.
Move near a knot to separate its timelines
Software usually preserves one history.WorldBraid preserves the worlds that could have been.
Conventional systems overwrite state as actions occur. An order ships, a balance changes, a permission is revoked, and the moment before is gone.
WorldBraid keeps every consequential state as an immutable version of the world. Each one can be reopened, inspected and continued as its own timeline, while the Origin stays untouched.
Your software already describes a world.
Its databases contain entities. Its APIs define possible actions. Its permissions determine who can act. Its traces reveal how state changes over time. WorldBraid compiles those fragments into a world agents can inhabit.
Diagram: OpenAPI, SQL, MCP, documentation, traces and rules converge into the Loom, and a single executable WorldSpec emerges.
In
Specifications, schemas, tool definitions, prose documentation, business rules, representative traces and optional application code.
Compiled
Entities, relationships, actions, permissions, resources, transition rules, scheduled events, actors and constraints.
Kept visible
Areas the inputs do not explain are marked as uncertain rather than filled in silently.
Connect the artifacts that describe your system. WorldBraid identifies entities, relationships, actions, constraints and uncertainty.
Create an Origin: an immutable starting world shared by every experiment, agent or possible future.
Agents interact with the world through real action contracts. Every consequential decision creates a new Thread.
Time advances. People react. Events arrive. Resources move. Each world develops its own history.
Revisit any Knot, introduce another decision and continue a different future.
See where worlds diverged, which actions mattered and what assumptions shaped each outcome.
One shortage. Three futures.
A small fictional world: a parts manufacturer facing a supply shortage. Choose an action at the Origin, advance time, then rewind and choose again. Nothing here is scored. Each Thread simply shows what it cost and what it created.
Origin: Inventory is low. Demand is rising. Three contracts are at risk.
Choose an action at the Origin
World state
History of this Thread
The Origin has no history yet. Every Thread begins from this exact state.
Assumptions shaping it
Illustrative world with invented figures, written for this page. It shows how a Braid is explored; it is not output from a production model.
Give every agent the same beginning. See what each one creates.
Different models, policies or human-agent teams can receive identical copies of the Origin. WorldBraid records how their decisions generate structurally different worlds.
Agent A
Consolidates control.
Decisions route through a single actor. Many possible Threads collapse early into one narrow, fast-moving history.
Agent B
Delegates decisions.
Authority spreads across many actors. The world fans out into small local branches that rarely rejoin.
Agent C
Negotiates with external actors.
The Thread keeps leaving the system, reaching suppliers and partners, and returning to a world those actors have changed.
The point is not to rank these agents with one score. Each Braid is kept whole, so the shape of the world an agent produced can be read, questioned and compared.
A possibility is not a prophecy.
WorldBraid does not claim to know the future. Every world is conditional on its initial state, transition rules and assumptions. Observed facts remain separate from inferred behavior, and uncertainty travels with every Thread.
OBSERVED
Seen in real traces or data. Recorded, not assumed.
DECLARED
Written down by people: specs, permissions, business rules.
INFERRED
Estimated from evidence, with the evidence attached.
GENERATED
Plausible behavior produced to fill a gap, marked as such.
UNKNOWN
Not explained by any input. Left open and carried forward.
The provenance of every consequential world rule stays inspectable, so any future can be traced to what was seen, what was written and what was guessed.
Six kinds of questions a Braid can hold.
Operational digital twins
Explore alternative futures for software-defined operations.
Fork a world compiled from the systems that already run your operations, and follow an outage, a policy change or a staffing gap forward without touching production.
Multi-agent societies
Study cooperation, conflict, institutions and emergent behavior.
Populate a world with many actors and watch norms, alliances and bottlenecks form, then rewind to the Knot where they first appeared.
Strategic exploration
Follow the consequences of decisions without pretending one outcome is inevitable.
Hold several futures side by side, each carrying the assumptions it depends on, so a discussion can be about conditions rather than forecasts.
Agent planning
Allow an agent to explore futures before selecting an action in the real system.
An agent can branch a compiled copy of its environment, inspect a few Threads ahead, and bring back only the decision it chooses to make.
Research
Study how model choice, memory and incentives produce different worlds.
Change one variable, keep the Origin fixed, and compare the structure of the Braids that result rather than a single benchmark number.
Rules you wrote. Possibilities it generated. Both labeled.
A WorldSpec is the declarative description of a world. Declared rules sit beside inferred and generated behavior, and every line keeps its source.
observed
declared
inferred
generated
supply-shortage.world
The future is not a line.It is a braid.
Begin with an Origin
Build a world. Place an agent inside it. Follow what happens next.