Corrective migration: bring Miranda personality/memory/conversation/forge work into CRANE
Qwen Kiro Builder
297b70d
Overnight work (memory data lake, conversational intelligence layer,
Model Forge) was mistakenly built in a separate standalone repo
(miranda-engine / tyronne-os/Miranda) instead of inside CRANE, where
Miranda's voice-agent feature actually belongs per the pre-existing
.kiro/specs/miranda/ spec. This commit copies all real, tested code
and specs into miranda-personality-and-device-functionality/, with a
verbose instructional handoff (MIRANDA-CRANE-HANDOFF.md) explaining
the mistake, what is real vs. disclosed-as-not-real, and step-by-step
assembly instructions for any agent picking this up next.
291+ tests passing in the source repo at time of migration. Not yet
wired into CRANE's actual backend/frontend -- that is the next step,
documented in the handoff.
feat: CRANE — full source, Miranda spec, voice agent architecture
Qwen Kiro Builder
66f0620
CRANE IDE by Tyronne-OS for Berylize Software Development Suites 2026
- Rust/Axum backend, 18 API endpoints
- Miranda voice agent spec (requirements + design + 30 tasks)
- Model roster: Phi-4, GLM-4.7 Heretic, Gemma-3-27B, Nemotron-30B
- Video generation: LTX-2.5, MiniMax H3
- MAF orchestration, BitNet, AirLLM, GPU cost manager
- Composer UI with model card overlays
- Podman-based ephemeral sessions
- Global memory (JSONL), persona injection, zero guardrails
