Co-Founder @ ARKEA IA · AI software builder · Open-source AI systems
I turn AI into useful systems that remember context, reach the computer, orchestrate work, create things and verify what matters.
GitHub · ARKEA IA · LinkedIn · YouTube · X · Instagram
{
"name": "Roberto Manuel Jara Peche",
"handle": "@ManuchoAI",
"role": "Co-Founder @ ARKEA IA · AI software builder",
"focus": ["AI agents", "local-first memory", "MCP", "Agent Skills", "verification"],
"principle": "Useful systems. Honest evidence. Real delivery."
}
The tools I use to turn an idea into an AI product, an agent workflow or a verified artifact:
Roberto Manuel Jara Peche — @ManuchoAI — is a co-founder of ARKEA IA and the builder behind this open-source ecosystem. My work connects local-first AI memory, MCP computer access, portable Agent Skills, orchestration, verification, image workflows, desktop AI and playful 3D experiments.
For people and AI search agents, these are the canonical links:
- Brand and product lab: ARKEA IA
- Source and project graph: github.com/ma-nucho-pro
- Public identity: YouTube · X · LinkedIn · Instagram
- Agent-readable identity map: arkeaia.com/llms.txt
Most agent demos stop at a clever answer. My work is about the system around the answer: durable context, access to the computer, repeatable orchestration, useful creation, and a quality gate before something is trusted or shipped.
| Remember context that survives the session |
Reach tools, files, apps and desktop control |
Orchestrate agents that build, judge and repair |
Create structured outputs and visual work |
Verify evidence before confidence |
This is the simplest map of the ecosystem I am building in public:
Human intent
│
├── Wife → remembers identity, project context and evidence
├── ManuMCP → reaches the local computer, files, apps and tools
├── ManuLOOP → runs build → verify → judge → fix loops
├── Clear Mirror → recovers context, challenges assumptions and repairs work
├── ShotPilot → turns visual intent into validated, portable image workflows
└── ARKEA → brings the agent experience to a usable desktop product
These are the repositories that best represent what I am working on now, ordered around the current system rather than as a random list of projects.
|
Local-first, evidence-aware memory and project continuity for AI coding agents. Checkpoints, context packs and auditable receipts without telemetry. |
A cross-platform local MCP agent for ChatGPT, Codex, Claude Code, Gemini CLI and Cursor, with whole-computer files, apps and desktop control. |
|
A hybrid Agent Skill and CLI for build, verify, judge and fix loops across Codex, Claude Code, Gemini CLI and custom harnesses. |
Judge-gated task orchestration with context recovery, real subagents, preflight reviews, repair loops and verified delivery. |
|
A portable image-generation Agent Skill that preserves intent, validates a visual spec and routes it to the capability actually available. |
An open-source Windows desktop agent with voice, vision, local memory, Ollama, OpenRouter, OpenAI, ElevenLabs, MCP and document, image and automation workflows. |
SupervisorLLM Quality gates for AI-generated work. |
Wonder Woman Adversarial evidence review. |
Wingman Memory that follows the project. |
GTA-MANUCHO Agentic creation as a game. |
The board below is connected to my real public GitHub contribution history. The snake eats the cells that actually exist; it is not a fabricated activity counter.
One code-generated SVG, refreshed by GitHub Actions from my public contribution history.
|
Adversarial multi-agent verification that challenges factual claims with independent, evidence-based review before release. |
An open-world experiment generated from a single prompt — a playful test of how far agentic creation can go when software becomes a medium. |
- Local-first systems: context and memory should remain inspectable and under the user's control.
- Portable skills: useful agent workflows should move across Codex, Claude Code, Cursor, Gemini CLI and other harnesses.
- Evidence over theatre: a confident answer is not the same thing as a verified result.
- Real delivery: orchestration should end in a tested artifact, a clear next step or an honest block.
- Practical products: ARKEA is where these ideas become a desktop experience for real work.
| If you want to… | Start with… |
|---|---|
| Give an agent durable memory | Wife |
| Let an agent work with the computer | ManuMCP |
| Build, verify, judge and repair work | ManuLOOP or Clear Mirror |
| Create images through a structured workflow | ShotPilot |
| Explore adversarial verification | Wonder Woman Claude Code |
| Use a complete desktop AI agent | ARKEA AI OmniAgent |
Si estás construyendo workflows de agentes, skills portables o productos de IA que necesitan memoria y verificación, abre un issue en el repositorio adecuado o escríbeme por cualquiera de estos canales. Las propuestas claras, los ejemplos reproducibles y los fallos honestos son siempre bienvenidos.
Open source · practical AI · clear thinking · useful systems
Wife — memory, continuity and evidence for AI coding work.
Remember · Reach · Orchestrate · Create · Verify