Open source
codex-agent-mem v1.0.1
Local-first memory for agent continuity. It preserves operational context across sessions, reducing repeated context reconstruction through local and reproducible context packs without depending on cloud memory.
We build tools for an internet of agents.
Internal R&D
Visual Systems Lab.
Applied research for operational, local and verifiable AI systems.
Visual Systems Lab is Visual AI Media's internal R&D environment. We design tools and criteria for agents and AI systems to work with continuity, evidence and control inside real environments.
What the Lab is
The Lab connects technical research with practical implementation.
It is not a separate commercial brand or a product catalog. It is the space where we validate approaches, develop our own tools and turn operational AI problems into publishable systems, papers or research lines.
The Lab publishes only what has sufficient evidence. Everything else feeds research lines until there is a verifiable baseline or a clear reason to make it public.
Published
Open source systems with public evidence.
Open source
clean-process-ended v0.7.3
Verifiable session closure for agents. It reviews local processes in dry-run mode, identifies cleanup candidates and generates operational evidence before intervening in the environment.
Joint workflow
Memory + operational closure
Both systems work independently or together. codex-agent-mem sustains continuity and operational memory; clean-process-ended helps close sessions with process review and receipts.
Research focus
AI systems with memory, evidence, control and real adoption.
1
Local operations for agents: continuity, session closure and reproducible evidence.
2
Local contracts between agents and tools, with capabilities and limits declared in a verifiable way.
3
Private AI layers for organizations: operational memory, automation and their own data policies.
4
We create systems with operational governance so AI can be used, audited and adapted by human teams and practical work environments.
Lab papers
Short technical texts to document decisions and lessons learned.
Paper
View docs
Local-first memory for agents
Why repeated context reconstruction is an operational problem and how to approach it with verifiable local memory.
Paper
GitHub
Verifiable session closure
Dry-run, receipts and human control as criteria for closing agent work without leaving local processes disordered.
Research
Research line
Local agent-tool contracts
Contract-first interfaces for local tools to declare capabilities, limits and diagnostics without promising total control of the environment.
Research
Research line
Private AI layers
Internal AI environments adapted to each organization, with operational memory, automation, control criteria and their own data policies.
Process
From research to public evidence.
A research line can emerge from a technical question or empirical need, become a paper and, when there is a verifiable baseline, be published as an open source system or feed Visual AI Media implementations.
1
Research line
2
Technical paper
3
Public baseline
4
Applied implementation
Contact