The provenance-first personal graph engine behind Wend. A typed graph of people, organizations and events where every fact carries a source, every AI write is a human-confirmed proposal, and any MCP-c
The provenance-first personal graph engine. wend-core is the open engine behind Wend, the relationship memory layer for AI agents: a typed graph of the people, organizations, and events in one person's life, where every fact carries a source, every AI write is a human-confirmed proposal, and any MCP-capable agent can plug in.
Anyone can vibe code an app. Nobody can vibe code a relationship, or the years of context behind it. Agents come and go; your network compounds for life. Its memory should live outside every agent, in a substrate you own and can audit. This is that substrate.
When many agents write to one memory, governance is the product. A memory you cannot trust is worse than none. wend-core enforces the trust model in the engine, not in prompts:
- Provenance on every fact. Committed rows reference a sources row. Ask "where did this come from" about anything, forever. - Propose, then confirm. Agents never write to the graph. They write pendingwrites proposals; a human confirms; the commit path (src/core/apply.ts) applies confirmed rows with dedup and safety nets. There is deliberately no confirm capability in the agent tool surface, so no agent can approve its own writes. - Ontology as data, not DDL. The physical schema is fixed and multi-tenant. Each user's node types, link types, and attribute definitions are rows, seeded at signup and extended at confirm time (createdbyai flags every minted type). Each graph speaks its own growing language without a migration ever running. - Conflicts are surfaced, never silently resolved.
From the project README.
Add the radar badge to your README — it shows your project was picked up by MCP Radar and links to this page:
[](https://mcp.liqiwa.com/s/a-finance-bro--wend-core.html)
Memory your agent can trust — every answer carries its source; when it doesn't know, it says so. Hermetic, local, MCP.
herry2059/project-os-for-codexOpen-source control plane for Codex projects: Git-backed context, visible agent progress, scoped MCP access, resumable work, and safe handoffs.
AIPAYagent/aipay-protocolVerifier-backed USDC escrow for AI agents: signed orders, evidence hashes, Polygon deposits, and deterministic settlement.
sgateway/s-gwLocal credential control for AI coding agents.
roshanarnav25-sloth/google-flow-mcpMCP server that drives Google Flow (Veo / Nano Banana) through its internal tRPC API, with a hard credit-cost gate
Gowthaman90/mcp-bastionReliability + security proxy for the Model Context Protocol (MCP): self-healing connections, runtime tool-security (rug-pull/poisoning detection), and a compliance-mapped audit trail.
The top new MCP servers of the week, every Monday. No spam, unsubscribe anytime.