Overview

Nika OS is a semantic operating system: an agentic kernel orchestrating swarms of Claude Code pods, three-tier memory, and evolutionary strategy optimization.

Nika OS — Russian matryoshka doll with straw hat, BCUB3 isometric cube logo at heart

Harness custom constantly evolving

Multi-pod orchestration · Recursive · Antifragile

What Nika OS is

Nika OS is the agentic runtime BCUB3 uses to automate the intellectual work of the industrial shop floor. It is not a chatbot. It is not a bag of scripts. It is a system that:

  • Orchestrates swarms of agents (Claude Code instances, called “pods”) through a real-time inter-process communication layer.
  • Persists knowledge across three memory tiers: working memory, episodic memory, semantic memory.
  • Evolves its own strategies through prompt-optimization tournaments (inspired by the GEPA research, 2025) — without ever mutating its kernel.
  • Drives probabilistic industrial controllers (the KTW B1 and B2 patents) on behalf of customer plants.

The “OS” terminology is deliberate: we treat agents as processes, memory as a filesystem, IPC as a system bus, and hooks as POSIX signals.

Why this design

Industry is not chatbot territory. A quality-control station receives 2,000 measurements a day, a machine setpoint shifts every batch, an operator changes shift every 8 hours. A naive assistant that restarts from scratch each session has no chance. The system must:

  1. Hold over time — memory beyond a single context window.
  2. Separate kernel from harness — engraved invariants, mutable parameters.
  3. Learn from stress — antifragility in Taleb’s sense, not just robustness.
  4. Trace every action — auditability by default, append-only JSONL trail.

Architecture in one picture

                      ┌─────────────────────┐
        user ───────▶ │   Alpha (kernel)    │ ◀── lifecycle hooks
                      │   orchestrator      │     (SessionStart,
                      └──────────┬──────────┘      PostToolUse, Stop)

                  ┌──────────────┼──────────────┐
                  ▼              ▼              ▼
              ┌───────┐      ┌───────┐      ┌───────┐
              │ Pod A │      │ Pod B │      │ Pod C │   Claude Code
              └───┬───┘      └───┬───┘      └───┬───┘   instances
                  │              │              │
        ┌─────────┴──────────────┴──────────────┴─────────┐
        │                                                 │
        ▼                                                 ▼
  ┌──────────────┐                              ┌───────────────────┐
  │  Semantic    │                              │   IPC bus         │
  │  memory      │                              │  (Redis Streams + │
  │  (Qdrant)    │                              │   JSONL bus)      │
  └──────────────┘                              └───────────────────┘

Who this site is for

This documentation describes the internals of Nika OS for three audiences:

  • AI/data engineers who want to understand how we structure multi-agent systems in production.
  • Researchers interested in the practical application of recent techniques (GEPA, prompt evolution, Kelly–Taguchi–Weibull control).
  • BCUB3 industrial customers who want to know what actually runs behind the deliverables we hand over.

Status

Nika OS is an active R&D project. The kernel is not open source at this time — it is a differentiation primitive for BCUB3. Utility components (skills, MCP servers, tutorials, neural activation functions) are progressively published at github.com/Powwpol under MIT.

This documentation reflects the state as of May 22, 2026.