JoseMunoz X , opens in a new tab
Jose Munoz in a dark editorial studio

I build systems that remember.

I’m Jose Munoz. I build memory, agent workflows, and local AI infrastructure by making the tools I want to use.

Explore the work X / @lukeNukemAI
Scroll to enter
Portrait of Jose Munoz Jose Munoz / builder

Curiosity,
made operational.

By day, practical operations. After hours, agents, memory, and local infrastructure. I learn AI by making systems I can actually use.

A question about whether an agent could remember turned Python lessons into Mnemo. A desire to own the runtime led to local models, Docker, and a private machine mesh. Each build teaches the next one.

The point is not another chat window. It is useful continuity: systems that retain context, divide work deliberately, and leave behind an artifact a person can inspect.

  1. 01 Local models

    Running experiments on machines I control.

  2. 02 Multi-agent work

    Giving models roles, handoffs, and checks.

  3. 03 Mnemo

    Connecting vector recall to graph context.

Built to hold context.

Memory, orchestration, and local control—four connected experiments shaped by real use.

01

Mnemo / vector + graph memory

Memory is more than retrieval.

Mnemo gives agents per-user continuity across conversations. Semantic recall finds the relevant memory; graph relationships preserve how people, projects, events, and decisions connect.

  • Vector recall
  • Knowledge graph
  • Memory dashboard
Open Mnemo
02

Harness / local GLM runtime

The model meets the command line.

A terminal-first agent runtime and shell copilot for local GLM models—with tools and working context close to the place where the work already happens.

03

Multi-agent workflows

Many agents. One accountable outcome.

Codex, Claude, Grok, Kimi, and Pi take deliberate roles in research, writing, implementation, and review. The human stays responsible for the goal and what ships.

See The Bergenline Beat
04

Local AI stack

Keep the runtime close.

vLLM and Ollama serve models, Docker contains the services, and Tailscale links the machines—a private foundation for learning without giving up control.

What I’m
learning.

Graph thinking: how relationships can make memory, retrieval, and reasoning more useful.

Neo4j EDU / Cypher course 10 practical lessons

Learn Cypher by following the graph.

I’m learning connected data by teaching it back. This course moves from graph fundamentals to writing Cypher, finding paths, and reasoning about relationships.

Start the Neo4j course
MATCH WHERE CREATE PATH MATCH (learner)-[:FOLLOWS]->(graph)
  1. 01Cypher fundamentals
  2. 02Creating & updating data
  3. 03Patterns & path finding
  4. 04Aggregation & grouping

Local first.
Agent ready.

A practical stack for systems that can recall, coordinate, and keep running.

Amber mineral memory object surrounded by fine orbital rings

Neural Cache / featured venture

AI systems,
built for the work.

Neural Cache turns agent research, private memory, and local AI infrastructure into practical systems a team can operate.

Visit neuralcache.io

Let’s build
something
useful.

Follow the experiments, work in progress, and what I’m learning next.

dim.cuba6112.tech