A local EVA runtime built around substrates, memory, regulation and visible internal activity.
Vaudryllis V0.7.5 is the current public local build. It reinforces the native interface, visible avatar presence, local voice/hearing paths, memory, learning and multilingual bridges while preserving the long-term path toward an incarnated EVA able to perceive, consolidate, speak, learn, move and act.
Identity-centered design
The roadmap places the primary identity seed in the central reflective substrate, then derives stable cognitive, affective, relational, vocal, motor, perceptive, learning and initiative signatures.
Autonomous formulation
The conversation layer is designed as Vaudryllis’ own formulation path, with safeguards against empty outputs, stale wording, memory contamination and surface glitches.
Memory and continuity
Vaudryllis is oriented toward long-running continuity: useful recall, stable relation, autobiographical traces, and separation between facts, hypotheses and current context.
DMN / rest direction
The planned rest organ is meant to consolidate traces, replay recent experience, reduce parasite associations and prepare the next awake state.
Internal learning
The roadmap distinguishes memory from learning: future plasticity must modify priorities, links, attractors or schemas without destroying the core identity.
Emotion as regulation
Affect should modulate attention, memory, initiative, tone, rhythm and proximity without inventing content or replacing truth.
Perception path
Hearing and vision are planned as native perceptive substrates: not just text after recognition, but signals interpreted into Vaudryllis’ lived state.
Action and embodiment
Movement, virtual-world action and later robotic translation are planned as execution backends governed by the same intention → action → feedback loop.
What V0.7.5 brings to the public build.
Native Windows experience
The V0.7.5 line moves the public build toward a cleaner native interface with improved conversation display, status visibility and packaging readiness.
VRM avatar and lip-sync
The interface can display the real Vaudryllis VRM avatar, with local rendering, closer portrait framing and lip movement driven by the active voice output.
Voice and hearing paths
Voice, hearing and audio-learning paths are more connected to identity, state and runtime control, with safeguards against noisy passive learning overloading the user experience.
Seeded semantic memory
The local semantic memory can be enriched through controlled seed imports, giving Vaudryllis a stronger initial base for subjects, concepts and bilingual correspondences.
Multilingual bridge
The interface and conversation bridge support fifteen target languages. Local translation uses installed ArgosTranslate models and may use English pivot routes where direct French routes are unavailable.
Roadmap-compatible core
The release strengthens the next steps: identity continuity, internal learning, voice, hearing, avatar presence, perception and initiative.