UNPKG

oneie

Version:

Build apps, websites, and AI agents in English. Zero-interaction setup for AI agents (Claude Code, Cursor, Windsurf). Download to your computer, run in the cloud, deploy to the edge. Open source and free forever.

77 lines (58 loc) 3.03 kB
--- title: Cycle dimension: connections category: cycle.md tags: agent, ai, claude, connections, cycle, knowledge, ontology related_dimensions: events, knowledge, people, things scope: global created: 2025-11-03 updated: 2025-11-03 version: 1.0.0 ai_context: | This document is part of the connections dimension in the cycle.md category. Location: one/connections/cycle.md Purpose: Documents connection-driven cycle workflow Related dimensions: events, knowledge, people, things For AI agents: Read this to understand cycle. --- # Connection-Driven Cycle Workflow Purpose: Show how every integration in `one/connections/*` maps to ontology connections and how to use those connections to drive retrieval and tool routing for cycle. ## 1) Scope via Connections - Seed: `actorId` (creator/user/agent) + `orgId` via `member_of`. - Expand content scope by traversing connections: - Ownership: `owns`, `created_by` - Content graph: `authored`, `part_of`, `references`, `published_to` - Teams/org assets: org → resources via `owns` Convex query (scope): ```typescript const owned = await ctx.db .query('connections') .withIndex('from_type', q => q.eq('fromEntityId', actorId).eq('relationshipType', 'owns')) .collect(); const scopeThingIds = new Set(owned.map(c => c.toEntityId)); ``` ## 2) Retrieve Knowledge (Vectors) - Filter knowledge by scope and type: `knowledgeType='chunk'`, `sourceThingId in scope`. - Vector search (ANN) on `knowledge.embedding` with k=10–20. - Hybrid rank by knowledge labels (industry/topic/format) and connection `strength`. ## 3) Determine Capabilities via Connections - External agents/workflows are things; connect via: - `communicated` (A2A/ACP/AG‑UI) - `delegated` (n8n/MCP/tasks) - Read connection `metadata` for protocol, endpoint, allowed tools, quotas. ## 4) Act and Log Events - When sending/receiving: `communication_event` with `metadata.protocol` and `messageType`. - When delegating: `task_event` with `metadata.action` and protocol. - When paying/fulfilling: `payment_event`, `commerce_event`, `fulfilled`, `approved`. ## 5) Learn Back into Knowledge - Upsert outputs as `knowledge` items (labels/chunks) and link via `thingKnowledge`. - Update `strength` or add new `references` connections if content cites other things. ## 6) Governance & Permissions - Always enforce org scoping through connections; only traverse/return items reachable by permitted relationships in the org. - Approvals are represented as `approved` connections or `notification_event` + `approved` edge to the resource. ## Canonical Mappings by File - A2A / ACP / AG‑UI: `communicated` connections + `communication_event`. - X402 / ACPayments / AP2: `transacted` connections + `payment_event`/`commerce_event`; `fulfilled` + `approved` as needed. - N8N / MCP / Service Layer: `delegated` connections + `task_event`. - Multitenant / KYC: `member_of`, `owns`, `created_by`. - Vectors: knowledge chunks linked via `thingKnowledge`; retrieval guided by connections.