Practice what you preach. I coach teams on shipping AI products; these are the products I ship myself. Each one solves a problem I actually had — cost control on agent workflows, personal knowledge memory, forecasting from sparse data, automated ops. Some are open source, some are in pilot, all are running in production.
Agentic memory for teams running more than one agent. Agents can only append raw notes; a separate compiler is the only writer to the wiki, so safety comes from structure rather than trust. It detects contradictions between what different agents believe, resolves the ones it can with a confidence threshold, and escalates the rest to a human queue. Recall fires passively on every turn over a hybrid keyword-plus-vector index.
Agent-discoverable estimation service for uncertain-variable forecasting. Provides structured probability estimation via REST API and MCP server, powering both Krobar.ai and standalone agent workflows.
A provider-agnostic ideation function (multi-stance generation plus an expansion pass) that drops into any model client.
Krobot, Kromatic's public MCP server — the full innovation-coaching knowledgebase (frameworks, client engagements, methodologies) and the complete text of The Real Startup Book, one call away from any AI agent.
Monte Carlo simulation engine, exposed as an MCP server — runs probabilistic root/calculated-node graphs, solves distribution parameters, and does sensitivity/scenario analysis for agents reasoning under uncertainty.
Synthetic user panels for any artifact, run across multiple model providers to correct for self-preference bias. Returns the point a persona quits, not a warmth score.
A cost router for scheduled agent work. Cheap scripts run first, a small model classifies, and a frontier model runs only on items that actually need reasoning.
OSS release-readiness gate — deterministic pre-gates (Scorecard, REUSE, Gitleaks) plus a no-context reviewer panel built on panelist, rendering pass/warn/block verdicts.
Data-driven decision making for enterprise innovation portfolios — Monte Carlo simulations for financial modeling under uncertainty.
Personal Chief of Staff built on NanoClaw — manages email, calendar, bug-fixing, and the automated build pipeline. 47+ open-source skills across communication, scheduling, development, and infrastructure.