Qortex
Agents that remember what the organization knows.
A capable model with no memory starts every conversation from nothing. Qortex gives agents a place in an organization — a tenant, an organization, an environment, a set of skills they are allowed to use — and a layered memory that separates what an agent knows about itself, about the person it is working with, and about the institution it works inside.
For · Organizations deploying AI agents against internal knowledge and processes.
CAPABILITIES
What It Does
Layered Memory
Embodied, personal, externalized and institutionalized memory are distinct stores — what an agent is, who it is helping, what it has recorded, and what the organization holds.
Organizational Model
Tenants, organizations, humans, agents and environments are first-class — an agent has a place in a structure, not just an API key.
Scoped Skills
Capabilities are granted at platform, tenant or organization level, so what an agent can do is a deliberate decision.
Audit Log
A record of what an agent did and on whose behalf — the question that gets asked the first time an agent does something surprising.
Environments
Separate the contexts agents run in, so an experiment and a production process are not the same surface.
Provider Abstraction
The platform is not welded to a single model vendor.
HOW IT WORKS
Start To Running
- 01You define an organization and the humans and agents inside it.
- 02You grant each agent the skills it is allowed to use, scoped deliberately.
- 03As agents work, what they learn is written into the memory layer it belongs to.
- 04Later sessions resolve against that memory, so context carries across conversations and across people.
- 05Every action lands in the audit log, attributed to the agent and the person it acted for.
UNDER THE HOOD
Architecture
Qortex Visio
A linear chat transcript is a poor medium for building a mental model of a complex topic. Visio is an experiment in whether an AI can work in the surface where people already reason well — a spatial one.
It's an Obsidian plugin that turns the Canvas into a shared visual-spatial working memory between a person and a model. We're testing a hypothesis, not claiming a result about how cognition works — it's early, desktop-only, and actively evolving.
- Classifies a prompt or graph into a structural shape — hierarchy, pipeline, cycle, network, and others — instead of applying one generic auto-layout.
- Lays out and maintains native Obsidian Canvas nodes and edges, so the surface stays a real working document.
- Treats the canvas as shared memory between a person and the model, rather than a picture generated once and discarded.
Qortex Impromptu
People hand vague prompts to expensive agents, contractors and colleagues — then pay for the wrong thing to get built. Impromptu is a refinery that makes a prompt precise before you spend that round.
It runs as a sequence of refinement rounds on the same canvas as Visio. Each round diagnoses what is underspecified, resolves what it can, and asks only what it genuinely can't. Like Visio it's early, desktop-only, and a research bet — not a product.
- Each refinement round diagnoses what is underspecified and rewrites the prompt under fixed headings — Goal, Context, Deliverable, Constraints, Non-Goals.
- Gaps split into stated assumptions you can reject in one click and genuine questions that need an answer, rather than one vague block of clarifying text.
- Convergence is structural — the chain closes when no questions remain, never on a model self-claim — and every round lands as real notes joined by edges, an audit trail you can re-run.
EARLY CONVERSATIONS
Interested in Qortex?
Qortex is still early — we're talking with a small number of pilot partners. If it fits what you're building, we'd like to hear from you.