Claude · MCP · per-agent memory · knowledge graph · governance layer
A product built from my own agent operating system: a fleet of AI agents that runs an organization’s inbox, outreach, social, finance and research, under a governance layer where nothing ships without a human yes. Running in production, daily.
- A fleet of AI agents that run inbox, outreach, social, finance and research, with nothing shipping without a human yes.
- Propose, review, approve, execute, audit: a code-level governance layer, per-agent memory, and MCP tool access under scoped credentials.
- Two-channel inter-agent comms and skill dispatch; work compounds into a user-owned knowledge base structured for fine-tuning.
havenware.dev
CONFECOOP Federation — Governed AI Assistant
LLM · RAG · MCP · Neo4j · human-in-the-loop · eval dataset
The governed AI assistant for CONFECOOP, the national confederation for Portugal’s cooperatives and social-sector enterprises, serving its 85+ member organizations. A RAG assistant over the institution’s own archive that always cites its sources, with document classification kept honest by a human-review queue.
- A RAG assistant over an institution’s own archive: answers are grounded in a Neo4j knowledge graph and always cite their sources.
- Incoming documents are classified by an LLM with a human-review queue; approvals and corrections are captured as a labelled dataset for evaluation and fine-tuning.
- Agents reach the platform over an MCP server under scoped permissions; every action lands on a tamper-evident, hash-chained audit log.
MCP server · RAG assistant · approval-gated ledger
A financial back-office platform, in production, where AI agents propose payments, payroll and reconciliations and a human approves them, on an append-only ledger. Adopted across public and social-economy institutions.
- Financial-operations agents that propose payments, payroll and reconciliations, but only a human can approve.
- Exposes an MCP server so agents act under scoped permissions, on an append-only, replayable ledger.
- A RAG assistant answers questions from the organization’s own financial records.
sigo.so
Evaluation & Automation Pipelines
n8n · Windmill · eval loops · scraping · reconciliation
- Tapering-review evaluation loops that grade agent output and feed the results back into prompts and datasets.
- Automation pipelines (n8n, Windmill) for ingestion, reconciliation and reporting, with human handoff where an automated route isn’t safe.
