Ghostlight is the studio’s social-AI experiment: agents designed to model perspective, local context, interpersonal stance, and scene logic instead of defaulting to the usual praise-addicted chatbot goo. The repo sits somewhere between runtime experiment, story-lab toolchain, and character-simulation doctrine.
Started
01 May 2026
Last Public Push
06 May 2026
Stack
Python / agent tooling
Status
Active experimental system
Founding Idea
The founding idea is right there in the description: stop building sycophantic chatbots and start building social agents. That means explicit state, point of view, projection boundaries, lore grounding, and scene rules strong enough to let a character disagree, misread, posture, hedge, or just be wrong in a recognizable way.
Trajectory
Ghostlight starts as a persistence spine and a document migration. Then it narrows hard into Call of the Void authoring, agent-state schemas, dialogue projection contracts, lore grounding, and story-lab fixtures like Cold Wake. Very quickly after that, the repo turns outward again and starts wiring itself into the wider stack: CultNet contracts, CultCache persistence, and swarm-state coordination. So the trajectory is not “build a chatbot.” It is “build a social runtime that can survive contact with actual scenes.”
Ambition
The ambition is larger than the current repo slice, and the history admits it. There are already commits preserving open-world training ambitions while the runtime scope temporarily narrows to dialogue scenes. That is sane staging, not retreat. The endgame is a socially legible agent system that can be used for authored fiction, simulations, and maybe eventually the nastier edges of interactive worlds.
History Tells On Itself
2026-05-01Scaffold Ghostlight persistence spine2026-05-01Retarget Ghostlight toward Call of the Void authoring2026-05-01Define prompt projection contract2026-05-01Adopt Ink for branching scenes2026-05-06Migrate state spine toward cultcache-py2026-05-06Define the shared CultNet agent-state contract