Apache 2.0 · Federated · Opt-in

Knowledge can travel.
Control stays local.

MoE Libris lets sovereign AI instances exchange curated knowledge bundles through explicit trust relationships, automated pre-audit and human approval.

Not a central brain. A reviewed exchange between independent nodes.

Follow a knowledge bundle Read the protocol
  • Voluntary participation
  • Human approval
  • Local acceptance rules
01 Sender sovereignty Nothing leaves a node without its decision.
02 Hub accountability Every bundle passes a visible audit path.
03 Receiver sovereignty No approved bundle is forced into a graph.

The federation principle

A shared library without forced assimilation.

Each deployment remains independent. Libris coordinates exchange, provenance and review; it does not own participant knowledge graphs or decide their local truth.

Central knowledge platform

One authority controls the corpus

  • Mandatory central storage
  • Opaque acceptance rules
  • One trust boundary
  • Participants surrender context

MoE Libris federation

Independent nodes choose their peers

  • Voluntary bilateral relationships
  • Sender-side privacy scrubbing
  • Review before global graph merge
  • Receiver decides what to import

One bundle, six trust gates

Knowledge earns its path through the network.

  1. 01 Extract A node selects reusable facts and relationships.
  2. 02 Scrub PII and local secrets are removed before transmission.
  3. 03 Validate Schema, predicates and content pass pre-audit.
  4. 04 Review A human administrator approves or rejects the bundle.
  5. 05 Publish Approved triples enter the hub’s global graph.
  6. 06 Choose Peers pull and independently decide whether to merge.

Trust boundaries

Security claims with explicit limits.

Federation endpoints are designed for protected HTTPS/mTLS deployment. Payloads remain intentionally inspectable by the hub so pre-audit and human review can work; there is no end-to-end encryption through the hub.

Sender

Before data leaves

Privacy scrubber, bundle selection, JSON-LD validation and API-key-authenticated node identity.

Hub

Before knowledge merges

Predicate allowlist, PII scan, abuse controls, audit queue and human decision.

Receiver

Before local acceptance

Explicit peer trust, incremental pull and independent import policy.

Open federation protocol

Small surface. Auditable behavior.

FastAPI endpoints exchange versioned JSON-LD bundles. PostgreSQL records durable audit state, Neo4j stores approved relationships and Valkey holds rate and abuse-control state. TLS termination and client certificates belong at the deployment boundary.

FastAPI PostgreSQL Neo4j Valkey JSON-LD mTLS 

POST /v1/federation/push
GET  /v1/federation/pull?last_sync=...
POST /v1/federation/handshake

bundle
├─ provenance
├─ entities
├─ predicates
└─ relations

Repository-verified status

Stable core. Explicit research frontier.

Status derived from the current API, service implementation and project memory on 16 July 2026. Deployment readiness still depends on secrets, certificates, databases and operator policy.

IMPLEMENTED

Sovereign node ↔ hub exchange

Push, incremental pull, handshake requests, node approval and API-key authentication are implemented under /v1/federation.

IMPLEMENTED

Review before graph commit

Schema and size checks, sensitive-pattern scanning and a predicate allowlist run before a bundle reaches the PostgreSQL audit queue. Only an admin-approved entry is committed to Neo4j.

IMPLEMENTED

Abuse and admin controls

Valkey-backed rate limits and strikes, automatic or manual blocking, timing-safe admin authentication and guarded outbound verification are present.

NEXT

Semantic LLM triage

The third pre-audit stage is a defined extension point but currently returns “not run”. It must not be described as productive semantic validation.

NEXT

Contradiction detection

Comparing proposed triples with approved graph knowledge before merge remains planned work.

ROADMAP

Hub-to-hub federation

Registry caching exists, but operational trust and synchronisation between Libris hubs remain the highest-priority open topology work.

MoE ecosystem

Libris connects knowledge. It does not replace the core.

S

MoE Sovereign

Creates validated knowledge and uses imported context in adaptive AI workflows.

moe-sovereign.org ↗
L

MoE Libris

Audits and federates voluntary knowledge bundles between sovereign nodes.

You are here
C

MoE Codex

Adds catalogue, lineage, approvals and data-audit capabilities where required.

moe-codex.org ↗

Federation remains a choice

Share what is reusable.
Keep what must remain local.

Inspect the repository Deploy a hub