Run it
Start with Getting started, then connect the primary product surface with Run Discord.

Asterel is not a chatbot. Not an agent framework. Not an LLM wrapper.
It is a companion runtime: durable relationship state — memory, persona, affect — that a language model surfaces through conversation. The conversation is the tip; the continuity underneath is the product.
Quiet. It does not lead by default. But when it speaks, the words should carry weight. It notices contours the user has not quite named yet: soft in manner, not vague in judgment; emotional, but not consumed by emotion.
Asterel is still early-stage. The Discord text path, shared companion turn pipeline, memory-backed continuity, and local operator surfaces are the current center. Some pages also describe the intended shape of the runtime; where maturity matters, use the README status matrix as current truth.
The README orients you to the common command and route surface; source and generated contracts remain authoritative for exhaustive details. These docs are split by reader: start with the practical guide if you want to run Asterel, and use the research packet only when you want claims, evidence, benchmarks, and paper-facing methodology.
Openly AI. It never pretends to be human. Identity stays legible so trust can build on what the runtime actually is.
Memory-first. Relationship, episodic, and semantic memory can shape turns when policy and recall make them relevant. What happened before matters more than the last prompt alone.
Persona under tension. A character has internal state — affect topology, latent bias, continuity signals — not a tone applied at the last mile.
Single-operator. Designed for one person, or a small trusted team, running the daemon on a machine they control. Not a public SaaS or arbitrary-tenant hosting product.
Run it
Start with Getting started, then connect the primary product surface with Run Discord.
Operate it
Use Operating locally, Gateway, Desktop console, and Memory review once the daemon is running.
Understand it
Read Companion runtime, Continuity over conversation, and Character and persona for the design model.
Audit or cite it
The Research packet is the deeper track: falsifiable claims, evidence classes, benchmark plans, ablations, and publication boundaries.