v1.0 — Initial Release
Foundation
First MCP server with memory and cognition tools. Proved the concept.
v2.0 — Hardening Baseline
Safety + Scale
Phase 7 hardening. Modular MCP decomposition (97 tools, 13 domains). Health endpoint, audit response, contamination firewall.
v2.1-rc2 — Current
Pluggable Integrations + Packaging
Blender and Unity as optional modules. Standalone packaging (requirements.txt, systemd, install.sh). Instrumentation endpoint (/metrics).
Q4 2026
Public Launch
Open-source release on GitHub. MIT license. Public documentation, beta program, community onboarding.
v2.2 — Post-Launch
External Vector Store
When the ledger exceeds 1M records, add an external vector store (e.g., FAISS at scale, Qdrant, or pgvector).
v2.2 — Post-Launch
LLM Cost Circuit Breaker
Auto-throttle LLM calls when provider bills spike. Protect runaway agent loops.
v2.2 — Post-Launch
Durable Metrics Sink
Persist /metrics to disk for historical analysis. Default on in 2.2.
v2.2 — Post-Launch
Figma + Design Integrations
Add Figma via the blank integration template. Other design tools follow.
v3.0 — Future
Multi-Agent Memory Sharing
Allow authorized memory sharing between agents in a fleet. Cross-agent recall with provenance.
v3.0 — Future
Memory Marketplace
Share and discover pre-trained memory profiles. Community-built cognitive templates.
| Feature | Status |
|---|
| Blender integration (7 tools) | ✅ Shipped |
| Unity integration (5 tools) | ✅ Shipped |
| Graceful degradation (no app = informative message) | ✅ Shipped |
| Standalone packaging (requirements.txt, systemd, install.sh) | ✅ Shipped |
| Event taxonomy + per-tool counters on /metrics | ✅ Shipped |
| pytest suite (contamination, eval safety, dispatch, metrics) | ✅ Shipped |
| Blank integration template | ✅ Shipped |