MetaCRDT

A substrate for structured coordination, local‑first computation, and human–to-agent symbioic interface.

Why MetaCRDT?

MetaCRDT is a unified lattice of Conflict‑Free Replicated Data‑Types (CRDTs), a typed task/agent model, and a local‑first execution kernel. Together, these primitives let us express arbitrarily rich workflows—from compliance data‑rooms to generative group games—while preserving:

• →State Convergence — Strong eventual consistency across arbitrary network partitions.
• →Auditability — Verifiable provenance and cryptographic time‑stamping of every mutation.
• →Composability — Typed schemas that compile to SQLite, Postgres, and Cloudflare Durable Objects.
• →Agent‑Native Semantics — Deterministic slots where LLM agents can observe, propose, and merge changes alongside humans.

Ongoing Research Programmes

• datarooms.metacrdt.com

  Structured Workspaces for compliance, onboarding, and regulated operations.

  Datarooms provide a structured environment for managing compliance workflows, onboarding processes, and regulated operations. They ensure that all participants have access to the right information at the right time, with full auditability and provenance tracking.

  How can schema‑driven CRDTs guarantee field‑level provenance while supporting offline edits?

• groupchat.metacrdt.com

  Conversation Graphs that treat every message as a first‑class CRDT node.

  Our conversation graphs model treats each message as a first-class CRDT node, allowing for rich, structured conversations that can be forked, merged, and analyzed. This enables new forms of collaborative dialogue where participants can explore multiple conversation branches simultaneously.

  What does "eventual consensus" mean for ephemeral, high‑velocity dialogue?

• rabbithole.metacrdt.com

  Exploratory Knowledge Mining—LLM agents recursively traverse data‑rooms to surface latent structure.

  Rabbithole enables LLM agents to recursively traverse data-rooms, discovering and surfacing latent structure and connections. This approach to knowledge mining creates reproducible research trails that can be verified, extended, and shared.

  Can automated rabbit‑holing generate reproducible research trails without hallucination?

• dataverse.metacrdt.com

  Polyglot Data Federation across heterogenous stores (CSV ⇄ Graph ⇄ Parquet).

  Dataverse provides polyglot data federation capabilities, allowing seamless integration of heterogeneous data stores including CSV files, graph databases, and columnar formats like Parquet. This enables unified access to diverse data sources while preserving their native semantics.

  How do we reconcile multi‑model data into a single CRDT lattice without lossy transforms?

• threadquest.metacrdt.com

  Procedural Narrative Systems where an AI Dungeon Master orchestrates branching story graphs in group chat.

  ThreadQuest explores procedural narrative systems where AI Dungeon Masters orchestrate branching story graphs in group chats. This research examines how to maintain narrative coherence when multiple players fork timelines concurrently, creating rich interactive storytelling experiences.

  How can narrative coherence be enforced when multiple players fork timelines concurrently?

Architectural Stack

1. Effect‑Core Runtime – Pure functional, interruptible fibers scheduling deterministic computations.
2. CRDT Kernel – Op‑based, delta‑encoded structures (G‑Counter, MV‑Register, Last‑Writer‑Wins Set, custom λ‑CRDTs).
3. Agent Interface Layer – Canonical event log → embedding cache → reflection prompts → merge‑proposal objects.
4. Transport Adapters – Gateways for WebSocket, Durable Objects, Server‑Sent Events, and p2p WebRTC.
5. Type Generation Pipeline – Single source of truth in @effect/schema → TypeScript types, JSON Schema, SQL DDL, and OpenAPI.