arkalabs · Agentic learning We build agents
We build agents
that acquire experience.
arkalabs develops Cortex, an agentic learning system designed to connect lived situations, decisions and their outcomes, then bring back useful learnings at the right moment.
Situation
Decision
Outcome
Learning
Revised · the reason is kept
Our thesis
Retrieving is not learning.
A history can be re-read. An experience becomes useful when it changes the way the next situation is approached.
Cortex is designed to reconstruct that experience, derive revisable learnings from it, and keep the reasons that produced them.
What must last
Your work produces more than traces.
A decision only makes sense with the situation that preceded it and the outcome that followed. Cortex maintains that trajectory beyond sessions, tools and models.
Three sessions, three tools, three days.
One single problem runs through them.
9 days · 3 tools · trajectory preserved
And the decision keeps what holds it.
HypothesisResume concentrates the load Evidence42 tickets · 3 weeks OutcomeDeadline met · debt contained
01
Resume
Recover the real state of the work, with decisions made, known constraints and points still open.
02
Decide
Reuse evidence and reasons already established, without rebuilding the whole context.
03
Revise
Correct a learning when reality changes, without erasing what made it valid before.
Continuity
Your experience stays in your environment.
Start with Codex, continue with Claude or switch provider. Cortex keeps the trajectory of the work and prepares, for each situation, the relevant material for the engine in use.
- 1You work in your usual tools. Your conversations, decisions and outcomes stay tied to the work that produced them.
- 2Cortex reconstructs the trajectory. It connects lived situations beyond session and application boundaries.
- 3Learnings come back when they become useful. The model receives the material relevant to the present situation, with its sources and its limits.
Codexat first 
Claudethen + Other enginelater
Cortex Cortex captures the work, then prepares the material relevant to the next engine.
Your experience · on your machineunchanged
Episodes
Decisions & reasons
Revisable learnings
Usable continuity
The past does not come back as a mere history.
Cortex does not return everything that resembles the request. It prepares what can genuinely help understand and handle the present situation.
01
Sessions become episodes
One problem can span several days, tools and conversations without losing its trajectory.
Notion · MondayClaude · WednesdayLinear · Tuesday EpisodeBilling resume flow
02
Decisions keep their reasons
Hypotheses, evidence and constraints stay tied to the conclusion.
HypothesisResume concentrates the loadEvidence42 tickets · 3 weeksConstraintNon-negotiable deadline
03
Outcomes close the loop
What actually happened after the action completes the experience.
the experience completes
04
Contradictions stay visible
New information can limit, challenge or revoke a learning.
05
Learnings keep their scope
A rule valid in one context does not automatically become a general truth.
Context · support team LearningShip in batches
06
Recall stays situational
Cortex prepares the useful and admissible part of the past, not all the accumulated material.
Scope decisionEvidence · 42 ticketsMeasured outcome
Under your control
Your environment stays yours.
Cortex installs on your machine. Your workspaces, subjects, experiences and decisions stay under your control.
When you use an external model, the flows involved stay identifiable and configurable.
Your machine
Workspacessubjects, documents, threads
Episodes & decisionswith their reasons and evidence
Learnings & provenancerevisable, never erased
what leaves · prepared, identifiable, configurable
External model Receives the material prepared for the situation. Does not own your experience.
01
Local storage
Persistent cognition and work data stay on your machine.
02
Providers of your choice
Codex, Claude or another engine can be used without owning your experience.
03
Provenance preserved
A conclusion stays linked to the experiences and evidence that support it.
04
Revisable learnings
What is no longer valid can lose its influence without disappearing from the history.
05
Exportable data
Your working environment does not depend on a single model or provider.
Research and product
Advancing an agent through what it lives.
We develop the mechanisms, skills and evaluation protocols needed to verify that a past experience genuinely changes how a new situation is handled.
Cortex is the foundation of this work. Our goal is clear: after a year in an environment, an agent should no longer approach problems exactly as on day one.
Discover Cortex
Test Cortex in your real work.
Connect your tools, continue your usual tasks and let Cortex maintain the trajectory between situations, decisions and their outcomes.