A full conlang workbench — IPA chart, phonotactic legality heatmap, sound changes, paradigms, family trees. Persist languages to your account; share any one of them by read-only link.
Built on PHOIBLE co-occurrence data so inventory suggestions reflect what actually appears in human languages.
Try the public demo →Simulate plate motion over geological time. Rifts open, collide, subduct. Results feed terrain, hydrology, and biome generation downstream.
Per-tile invariants are visualised, not abstracted — you see plate areas, drift, and divergence regimes as you tune the simulation.
Coming with the waitlist →Hayba speaks Model Context Protocol. Any LLM with MCP support can author terrain, place actors, apply conventions, and orchestrate the full toolset directly in Unreal Engine.
Tools are organised by domain so you can scope what the agent sees — Scene, Editor, PCGEx, Zone Painter, Conventions.
Docs →Procedural settlements grow from upstream signals — tectonic uplift, river hydrology, biome viability, language families. Roads aren't drawn; they emerge from where people would actually need to walk.
Style guides parameterise vernacular: wall material, roof pitch, courtyard logic, façade rhythm. The same generator produces a North-African medina, an alpine timber village, or a delta market town depending on what the world demands of it.
Architecture pillar →Drive Hayba from a chat session, a desktop dashboard, or directly inside Unreal Engine 5. The state of your world lives in a single source of truth — every editor reads and writes the same canonical JSON.
Diff and version like code. Branch a continent, merge a language family, roll back a culture. The toolset assumes you'll iterate, not commit-and-pray.
See the pipeline →The full toolset is on GitHub. Schema definitions, simulation engines, the MCP server, the linguistics workbench — all of it. No telemetry, no required cloud account to use the local tools.
The hosted tier is the waitlist; the local tier is free. Bring your own LLM via MCP — Claude, GPT-4, Llama, anything that speaks the protocol.
GitHub →Hayba is built pillar-by-pillar in public. Each pillar lands as a working toolset before the next begins. You can track progress on the GitHub issues.
Specific worries about MCP, licensing, the hosted tier, and the open-source pieces.
The full local toolset is MIT-licensed and free forever. The hosted tier — cloud sync, share-links, the admin queue — is currently waitlist-gated while we shake out the first 50 users.
Only if you want agentic authoring. Hayba speaks MCP, so any client that speaks MCP works — Claude Desktop, Cursor, Continue, even local Llama runs through Ollama-MCP. The simulation, linguistics, and architecture tools all run without an LLM.
The web tools (linguistics workbench, dashboard) run in any modern browser. The MCP server is a small Rust binary. The UE5 plugin needs UE 5.4+. Tectonics and architecture simulations are CPU-friendly — no GPU required for headless runs.
Yes. Worlds you generate are yours. The toolset is MIT — no royalties, no per-seat fees, no commercial use restrictions on the output.
Because the hosted tier sits on a single self-hosted Supabase. 50 users is what one box comfortably runs. Once we're confident in the sync + share-link flow we'll open it up. Until then, joining the waitlist costs nothing and gets you in early.
Those are excellent reference tools — they help you write down a world you already know. Hayba is a generation tool — it helps you discover what the world is by simulating its causes. The two genuinely complement each other.