Offline-first AI text-world framework

An agent that explores a live linked-list world.

Text World Agent is a hackathon-ready, reusable framework: a swappable world config feeds a continuously updated room graph, while Ollama receives only the current-room facts needed to act.

Jump to live model

Uses local Ollama at 127.0.0.1:11434; a deterministic route planner keeps the session responsive if it is offline.

Judges' view

Five signals in one screen

Separate list and map pointers

The insertion-order next chain never replaces navigable north/south/east/west exits.

Frontier-aware navigation

Unvisited neighboring rooms are marked in every compact decision slice.

Visible self-correction

Wrong moves and lock conflicts fire correction events that shape the next prompt.

Reusable by config

Replace world data, items, locks, and a goal without changing the linked-list or updater engine.

Fast stateless calls

No history replay: only goal, room, objects, inventory, exits, visited names, and a useful hint are sent.

Event-driven at every turn

Movement, items, unlocks, corrections, and victory propagate to the reasoning log and UI immediately.

Reusable config · live linked lists · token-minimal agent loopOffline after local Ollama setup
Built with GenMB
Built with GenMB