Design
Technical foundations — the conceptual architecture before code.
These documents capture the design rationale, not the implementation. They explain why outheis works the way it does, drawing from operating system research and distributed systems theory.
Documents
Why OS Principles
The core insight: multi-agent AI systems face the same challenges that operating systems solved decades ago. Message passing, ownership semantics, fault isolation — these aren't arbitrary choices but proven solutions.
Systems Survey
A technical survey of operating systems and their applicable concepts:
- DragonFlyBSD — LWKT, per-CPU queues, ownership semantics
- Erlang/OTP — Actor model, supervision trees, "let it crash"
- seL4 — Capability-based access control
- Plan 9 — Everything as filesystem
- OpenBSD — Privilege separation, pledge/unveil
Architecture
The complete architecture specification: system structure, agent roles, ownership model, message protocol, dispatcher design, vault structure, configuration format.
Data Formats
Detailed specification of all data formats: message schema, configuration structure, memory storage, vault conventions, logging format.
Related Work
Survey of existing multi-agent frameworks and how outheis differs.
Agent Prompts
System prompt specifications for each agent: common principles, role-specific instructions, capability boundaries.
The Hybrid Memory Stack
Where code ends and LLM begins — the sharp boundary between deterministic structure and learned meaning. Why neither pure code nor pure LLM is sufficient, and how outheis divides responsibility between the two.
The Quality Threshold
Why model capability is not a gradient but a threshold — and why twenty small models cannot substitute for one capable one. Emergent abilities, the failure modes of below-threshold models, and what this means for outheis's architecture.