Repository navigation
Home
Orrin is an experimental symbolic-first cognitive runtime: a long-running Python agent with persistent goals, memory, control signals, host telemetry, metacognition, and an optional LLM tool layer.
It is designed to make agent behavior inspectable over time: what it attends to, what goals it pursues, how memory changes, how control signals bias action selection, and how the runtime responds to the machine it runs on.
Orrin is a research prototype, not production software and not a claim of sentience. Cognitive terms in this wiki refer to engineering mechanisms: memory, goals, attention, control signals, runtime state, and action selection.
- Getting Started — install, run, and read the UI in your first hour
- What is Orrin?
- The Cognitive Loop
- Quick Navigation Guide
- Glossary
- FAQ
- Symbolic-First Design
- Workspace and Ignition
- Binding and Workspace Writeback
- Thinking and Action Selection
- Action Selection and Bandit
- Learning and Adaptation
- Loop Phases: Detailed
- Memory System / Memory Subsystem Deep Dive
- Control Signals / Deep Dive / Module
- Symbolic Reasoning / Subsystem Deep Dive
- Goals: Executive vs. Daemon / Goals Daemon Subsystem
- Peers / Peers Subsystem
- Host Coupling / Supervisor
- LLM Integration
- Production and the Effect Ledger
- Quality Standard
- Expression Membrane
- Self-Code and Extension
- Native Language Model
- Existence and Lifecycle
- Configuration Reference
- Running with Docker
- Remote Access & Tunneling
- Desktop Packaging
- Security Model
- Benchmarks and Verification
- Troubleshooting
- Debugging Memory Issues
- Contributing
- Cognition Module
- Writing a Custom Cognitive Function
- Extending Symbolic Operations
- Adding a Custom Peer
- Tuning Control Signals
Source and maintenance: this wiki's source of truth is docs/wiki/ in the main repository
(originally based on the WIKI_STRUCTURE.md blueprint, now in docs/archive/). Update pages when
code or design changes; keep code pointers in each page.
- Cognitive Loop
- Workspace & Ignition
- Binding & Writeback
- Thinking & Action Selection
- Learning & Adaptation
- Effect Ledger
- Quality Standard
- Expression Membrane
- Self-Code
- Native Language Model
- Existence & Lifecycle