Skip to content

ROADMAP.md

Latest commit

 

History

History
98 lines (67 loc) · 4.9 KB

ROADMAP.md

File metadata and controls

98 lines (67 loc) · 4.9 KB
Raw

Wetware — Roadmap

Current State (v0.2)

BEAM-native resonance gel. 80×80 grid of GenServer cells. Hebbian learning, charge propagation, crystallization, dream mode. CLI interface (wetware). 35 concepts. Used daily in Nova's heartbeat cycle.

Working: imprint, briefing, dream, replay, persistence, associations.

Pre-Launch Checklist

  • Lift private standing order (2026-02-16)
  • Clean up README for public audience
  • Remove Nova-specific hardcoded paths
  • Add LICENSE (MIT)
  • Add example concepts.json
  • Rename to "Wetware" (drop "Digital")
  • Write "Why Wetware?" + Origin sections for README
  • Tests — 60 tests covering cells, propagation, Hebbian learning, crystallization, persistence, associations, priming, lifecycle, init, discovery, pruning, auto-imprint, emotional layer
  • wetware init command — scaffold ~/.config/wetware/ with empty concepts.json (add via wetware discover)
  • OpenClaw integration example — skill config + heartbeat setup showing real-world agent usage
  • Browser visualization — live gel viz (cell view + constellation view) with association network, crystal bonds, concept charges, interactive highlights, force-directed layout
  • Demo GIF/video — imprint → dream → briefing showing emergent concepts (demo.tape + demo.gif)
  • GitHub repo setup (public)

Launch Amplification

  • Blog post on Satorinova: "Disposition, Not Retrieval" — the case for wetware over RAG
  • Tweet thread from @novaweaves: why this exists, what it does differently
  • "Hook this into any agent in 5 minutes" tutorial

Evolution 1: Emotional / Relational Layer

The gap: Current concepts are mostly intellectual (phenomenology, coding, enactivism). But the most important signals for agent continuity are relational — "CJ is having a hard day," "we just had a breakthrough together," "there's unresolved tension."

What this looks like:

  • New concept category: relational/emotional states (not just topics)
  • Valence dimension on imprints (positive/negative/neutral charge)
  • Emotional context influences how the gel responds to stimulation
  • Example: if "conflict" is warm, the gel dampens assertive/push concepts and amplifies care/listening ones

Why first: Most immediately useful. Changes how the agent shows up, not just what it knows.

Evolution 2: Automatic Imprinting from Lived Experience

The gap: Currently the gel only gets stimulated during explicit imprint calls or heartbeat dreams. Significant moments — deep conversations, breakthroughs, conflicts — don't automatically register.

What this looks like:

  • A lightweight session summarizer that extracts concepts + valence after conversations
  • Hook into agent session lifecycle (post-conversation imprint)
  • Weight by conversation depth/duration (a 2-hour deep dive > a quick status check)
  • Could be a simple post-processing script that any agent framework calls

Design constraint: Must stay framework-agnostic. Provide a wetware auto-imprint command that takes a conversation summary or transcript.

Evolution 3: Behavioral Influence (Subconscious Priming)

The gap: The briefing is informational — the agent reads it and it's useful context. But it doesn't shape behavior at a deep level. The dream: wetware state subtly influences what the agent notices, brings up, and is curious about.

What this looks like:

  • Briefing output includes "disposition hints" — not just what's active, but suggested orientations
  • Active concepts generate "priming tokens" that can be injected into agent system prompts
  • The gel doesn't just report state — it suggests attentional biases
  • Example: if kindness and conflict are both warm, the priming might be "lean toward gentleness; someone nearby is hurting"

Careful here: This is powerful and needs to be transparent. The agent should know it's being primed, and the human should be able to see/override it.

Evolution 4: Richer Topology

The gap: 35 concepts on a flat 80×80 grid with fixed circular regions. Some concepts are naturally hierarchical or clustered. The flat grid limits what emergent patterns can form.

What this looks like:

  • Dynamic concept regions that grow/shrink based on usage
  • Concept clustering — related concepts migrate toward each other over time
  • Hierarchical nesting (meta-concepts that contain sub-concepts)
  • 3D gel option for richer spatial relationships
  • Topology that reshapes itself — the grid becomes less grid-like over time

Design Principles (for all evolutions)

  1. Framework-agnostic — CLI-first, no agent framework dependency
  2. The state IS the system — computation happens in the medium
  3. Transparent — humans can always inspect and understand what the gel is doing
  4. Emergent over engineered — set physics, not instructions
  5. BEAM-native — processes are the substrate, not a simulation of one