RiotSpark identity
The identity contract remains the authority for what RiotSpark is, independent of the active model and separate from remembered conversation material.
R2.0 ROADMAP ›01 / RIOTSPARK
A personality-first core
built to continue.
RiotSpark 2.0 is a self-owned, provider-neutral AI architecture designed to preserve identity, behaviour, governed memory and relationship continuity across different language models.
Language models provide reasoning. RiotSpark identity defines the system, Agnes Personality 1.0 governs how it behaves, and durable conversations plus governed memory preserve continuity under human control.
Agnes Personality 1.0
Agnes is the frozen behaviour authority. A context engine and behaviour compiler turn that personality into consistent expression without letting a model redefine it.
Governed memory and conversations
Durable conversations provide interaction history. Governed memory remains reference evidence that can be inspected, corrected, superseded, retired or deleted.
Portable human authority
SQLite-first storage, exports, isolated restore, safe mode and provider-neutral adapters keep the system recoverable and under owner control.
Why separate the identity from the model?
Identity and behaviour are different authorities
RiotSpark 2.0 separates the identity contract from Agnes Personality 1.0. The first defines what the system is; the second governs how it expresses itself. Neither is delegated to the active model.
Memory informs; it does not command
Governed memory is reference evidence, not behaviour authority. It may inform a response, but it cannot silently rewrite identity or personality. Conversation storage is also kept separate from governed memory.
Local-first, provider-neutral and recoverable
Ollama with Qwen3 1.7B is the primary local path and GPT is used as a reference path. Adapter boundaries, SQLite persistence, backup, isolated restore and safe mode reduce dependence on any single provider or machine.