This account is the work I build in the open.

I care about the same things in both worlds: governed memory, agents that follow rules and verify their own work, data sovereignty, and shipping fast without hand-waving.

coco is an open-source agentic-AI orchestration framework that turns Claude Code, Cursor, or Codex into a full engineering department: a 389-persona advisory board, 147 skills, and 277 commands, with local privacy and universal compatibility. Open-core — the framework is MIT; the Super Intelligence board is proprietary.

Original work, grouped by what it does.

Platform and memory

• CoCo Platform, a local-first control plane for AI coding agents that pairs a React UI (26+ surfaces) with a FastAPI backend, a governed memory architecture, and a self-evolving agent loop that halts at a human-approval gate (private).
• CoCo Hermes, the backend intelligence layer — API, RAG, a canonical entity spine (deduplicated to 0% fragmentation), a hierarchical memory tree, and a citation-gated retrieval router (private runtime).
• CoCo Memory and M0, a deduplicated entity spine plus a cross-tool operational thread that lets Claude Code and Cursor share one live context (private runtime).
• CoCo Router, sovereign model routing that dispatches every call by task, cost, and load and hard-blocks raw public-cloud egress for regulated content (private runtime).
• CoCo Connect, a one-command installer that wires CoCo into Claude Code and Cursor with idempotent, backed-up, health-checked connections (private runtime).

Governance and reasoning

• coco-loops, a governed autonomous-loop framework: a verifier that defaults to reject, an immutable eight-invariant constitution only a human can amend, a bounded-lifetime kill switch, and a promotion gate. Built with 114 passing tests and one verified live proof. Source-available.
• CoCo Superintelligence Teams, attributed decision panels across 9 expert-persona teams (389 cited real-world personas), used for real architecture and product decisions. Ships inside coco as the proprietary core of the open-core split.
• coco-fusion, a bring-your-own-key multi-model persona council, parked after an honest null-result evaluation against a single frontier model. Source-available.

Apps and tools

• Coco-Voice, a free, open-source, on-device speech-to-text desktop app.
• CoCo-PDF, a one-click desktop PDF editor (private).
• PM-Studio, a product-management deliverable and document studio.
• StoryDeck, a local-first, on-device sprint deck tool.
• PM-with-AI, a tool-agnostic methodology for working with AI as a product manager (private).
• Mira, a local-first short-form-video engine (English and Hindi, human in the loop). Private; will publish under AGPL-3.0.
• CoCo Graphify, a managed code-knowledge-graph; and CoCo Ads, a fully local launch-video skill adapted from the open-source brag skill (credited).

Co-author, with Leobardo Mora, of "Agentic Contracts: Smart, Bounded Specifications for Governable Stochastic Agents," the theoretical spine underneath CoCo Loops. Public listing pending clearance.

Alongside the original work above, this account keeps a curated set of forks I study to benchmark my own product decisions (model routers, agent harnesses, memory systems, and video tools). Those are other people's projects, kept for reference, and are not represented as mine.

Find me on LinkedIn.