Aeion Forge Agent Coordination Architecture

How does Aeion Forge actually coordinate multiple AI agents without them stepping on each other? The technical answer is four interconnected systems — session-handoff structured documents, a versioned + searchable knowledge graph, a per-agent memory store with selective-share semantics, and a conflict-resolution layer based on per-resource locks + claim-and-revoke patterns. This page walks through the architectural model so engineering teams building on top of Forge understand what they're getting.

Layer 1 — Session Handoff

`

Layer 2 — Knowledge Graph

`

Layer 3 — Agent Memory Synchronization

`

Layer 4 — Conflict Resolution

`

Per-Tenant Scaling Considerations

`

FAQ

Per-session ~50KB structured + ~10MB scratchpad. Per-tenant typical 1-10GB session storage.

Per-tenant 1M+ knowledge entries supported. Per-tenant pgvector HNSW scales linearly.

Per-tenant 1-5% of captured knowledge has conflict candidates. Per-tenant human-review process.

Per-handoff structured (not raw conversation). Typical handoff: ~5K tokens. Saves 50K+ tokens of context for next agent.

Per-tenant configurable. Per-Aeion-default file-level + issue-level. Per-tenant directory + module override.

Per-coordination 1-5% of agent time. Per-coordination tradeoff vs duplicate-work risk (10-30%).

Per-semantic-search sub-100ms. Per-text-search <50ms. Per-cross-collection <500ms.

Per-handoff schema-validated. Per-handoff completeness-score. Per-handoff next-agent feedback loop.

Per-tenant full export (JSONL). Per-tenant per-category export. Per-tenant migration to other systems.

Per-tenant immediate token revocation. Per-tenant rate-limit reset. Per-tenant audit.

No. Per-tenant isolated by construction. Per-tenant knowledge stays in tenant.

Per-Aeion-Aegis backup + PITR. Per-tenant session-handoff included. Per-tenant knowledge-graph included.

The coordination layer that makes AI agents work.