OPEN SOURCE
Agentic Loop Atlas
A field guide to agentic control loops
11 documented loops
Evidence
11 documented loop patterns with validation workflows, risk taxonomy, and implementation receipts.
- Agent systems
- Governance
- Verification
Selected workEvidence review · 2026
Mustafa Saraç · AI-native systems builder · Founder of NeuraByte Labs
Systems with evidence,not a timeline of claims.
Building agentic products across edge, cloud, and operations. Each featured item below points to a public repository, live surface, or first-party system note.
- Agentic workflow design
- Bounded loops, receipts, and approval gates.
- Human-governed operations
- Automation that keeps control visible.
- Edge inference
- On-device AI through practical tooling.
- Multi-agent systems
- Explicit roles, handoffs, and operating rhythms.
01 / Current proof
Featured systems
Four public artefacts that make the current systems-builder identity inspectable across operations, edge inference, and agent architecture.
SERVICE OFFER
AI Ops Pilot
Human-governed AI operations
7-day pilot model
Evidence
A public service model for introducing agent-assisted workflows without removing human approval and operational control.
- AI operations
- Human approval
- Small teams
OPEN SOURCE
cactus-rs
Rust tooling for on-device inference
Rust + edge AI
Evidence
Public Rust bindings and examples for running Cactus inference workflows on edge devices.
- Rust
- On-device AI
- FFI
SYSTEM NOTE
RICK / MORTY / SUMMER
A three-agent operating model
Strategy · execution · reporting
Evidence
A first-party architecture note documenting agent roles, cron jobs, handoffs, and working rhythms.
- Agent mesh
- Automation
- Operations
02 / Agentic work
Agent-assisted releases
Public artifacts shipped through a human-directed MORTY workflow, with ownership and evidence boundaries kept explicit.
MORTY-ASSISTED OPEN SOURCE
IDEA SCOUT / MORTY macOS Diagnostics
Small diagnostic tools shipped through an agent-assisted loop
7 public repos · 7 releases · MIT
Evidence
Mustafa’s IDEA SCOUT workflow with MORTY produced seven released utilities for inspecting disk usage, Spotlight, memory pressure, Continuity, USB-C signals, logs, and volume-ejection failures on macOS.
Product direction and safety boundaries by Mustafa; implementation and release workflow by MORTY / IDEA SCOUT. No production-use or adoption claim.
- Agent-assisted delivery
- macOS
- Diagnostics
- Open source
03 / Knowledge systems
Guides and frameworks
Public operating knowledge and framework prototypes, presented within their actual evidence boundaries.
GUIDE
Hermes Zero-to-Hero
A bilingual path into Hermes Agent
Evidence
A public Turkish and English visual guide for moving from chatbot use to inspect-plan-act agent operations.
- Hermes Agent
- TR / EN
- Operator guide
OPEN SOURCE PROTOTYPE
RUNE
Prompt lineage for repeatable agent workflows
Evidence
A public framework prototype exploring structured prompts, governance, and agent-cron lineage.
Presented as a framework prototype, without benchmark or certainty claims.
- Prompt systems
- Lineage
- Governance
04 / Technical exploration
Live technical prototypes
Exploratory prototypes and demonstrators — not production claims.
LIVE PROTOTYPE
Lithosphere
Procedural worlds in WebGPU
Evidence
A live browser prototype exploring real-time procedural rendering, materials, and interaction with WebGPU.
- WebGPU
- Procedural graphics
- React
LIVE DEMONSTRATOR
Silent-Ear
Predictive-maintenance concepts in Rust
Evidence
A public technical demonstrator exploring machine-condition signals with observable Rust interfaces.
A technical demonstrator, not a production or certified system.
- Rust
- Telemetry
- Observability
05 / NeuraByte Labs
AI experiments catalogue
A wider lab surface for public experiments that are useful to inspect but do not all belong in the featured proof set.
Historical record
Earlier product and interface work
The previous portfolio remains available as a separate archive. It records earlier product, interface, and open-source work without defining the current AI-native practice.
Next system
Building an agent-assisted operation?
Start with one bounded workflow, visible approvals, and evidence that survives the demo.