The best systems grow from simple rules that interact. We write constraints and initial conditions, then observe what forms. The architecture of BlackAgent emerged from thousands of iterations where the kernel's boundaries shaped the agent's behavior into something we did not prescribe.
Constraints are the parameters that define the search space. A token budget forces a memory system to prioritize. A clarification loop forces specificity before generation. A deterministic FSM forces safety independent of the model's judgment. Every constraint we impose is a design decision that makes the output space smaller and more useful.
Software that matters runs where people are and speaks how they speak. BlackAgent runs on a Raspberry Pi in Casablanca. Nadar speaks Moroccan Darija. Locality is a design philosophy that shapes everything from latency budgets to embedding models.
BlackAgent showed that reliability improves when autonomy is earned, but production stability improved even more once each complex task carried an explicit contract for scope, budget, tool limits, and escalation. The follow-up is to formalize which contract templates fit which task families.
Intent promotion works best when weak signals stay as memory until reinforced and promoted tasks remain linked to the originating memory. The next research pass is about tuning decay windows so reminders disappear when context changes, without cancelling useful long-horizon commitments too early.
Nadar performs better when it is proactive, especially for hazards and high-signal context, but there is a line where proactive behavior can become noisy. The follow-up is to define adaptive thresholds for intervention by environment type, user pace, and recent interruption patterns.
Cross-language recall is already strong in Nadar, especially across Darija, French, and English. The open question is how retrieval ranking changes when the same fact appears in mixed dialect forms and whether relationship memory should be weighted differently than factual memory.
SudoPrompt proved that forcing specificity before generation raises output quality. The next step is a lightweight scorecard that tracks whether each clarification round reduced retries, reduced variance, and improved first-pass usefulness across image, agent, and docs workflows.
Across SudoPrompt and BlackAgent, the strongest gains came from narrowing decision space. A useful follow-up is to catalog which constraints preserve creativity and which ones over-constrain outputs into repetition.
TRAJEKT's ring buffer depth controls visual character more than any synthesis parameter. BlackAgent's memory layers shape agent personality more than the model choice. Across both systems, the container mattered more than the contents. The follow-up is to study which structural decisions have outsized aesthetic or behavioral impact and whether those patterns generalize.
ALPS evolves Lisp-like programs to predict binary sequences using compression as fitness. Across 88 experiments, ablations showed that crossover, tournament selection, and a module system each had zero measurable impact. The engine's effectiveness comes from two sources: Lamarckian MLE parameter refit and operator injection. Everything else was removed.
The thesis studies a supply-chain threat model built on subliminal learning results from 2024 to 2026. Core claim: deterministic controls protect artifacts, while behavior in foundation models is distribution-shaped and can drift through channels that pass content-level screening.
BlackAgent starts constrained and earns broader access through a trust score that rises with successful task completion and drops with failures. Combined with contract-governed execution and a deterministic state machine for safety decisions, this architecture produced agents that are measurably more persistent, more efficient, and more predictable.
Emergent is an independent AI and software laboratory based in Grand Casablanca, Morocco, and a subsidiary of Blackwoods Creative. To our knowledge, the first of its kind in the country. We are a team of young minds driven by curiosity and an obsession with pushing limits. We exist to bridge the gap between humans and machines, making them work together in ways that didn't exist before.
We build autonomous systems, accessibility tools, developer infrastructure, and experimental research; then we release what we can. The work is the point. If something we build is useful, the code is there. If something we discover is interesting, the writing is there. We do things because we cannot; until we do.
casablanca, morocco