Skip to content

Pardesco/hypernovum

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

91 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Hypernovum — Agent Ops for Your Second Brain

A 3D IDE and agent-ops dashboard for Obsidian, built on Three.js. Hypernovum turns your vault — your second brain — into a living cyberpunk code city: visualize projects and their backlinks in 3D, dispatch AI coding agents (Claude Code, GPT Codex, Antigravity CLI) with full vault context, post research quests for agents to resolve, and watch your entire fleet work in real time.

Each project note becomes a building. Status maps to color, priority to height, category to district, vault backlinks to glowing Neural Link arcs. A central Neural Core pulses with activity as you work, Data Arteries flow to buildings when files change, and every active agent orbits its building as a colored orb.

Hypernovum Obsidian Plugin

Features

City Visualization

  • Bin-packed layout with category districts, block outlines, and drag handles for rearranging
  • Procedural architecture — each category gets a unique silhouette (helix towers, data shards, ziggurats, quant blades, hex hives, memory cores)
  • Cyberpunk shader system with procedural windows, decay dithering, and bloom post-processing
  • Smart labels with CSS2D rendering and leader lines
  • Hover tooltips showing status, priority, health, and tech stack

Second Brain & Agent Ops

  • Prepare vault for AI agents — one click writes a marker-fenced AGENTS.md at the vault root: project schema, live inventory, quest board, skills roster, and heartbeat protocol, so any CLI agent instantly understands your second brain
  • Quest board: a questions: list in project frontmatter renders as a floating gold quest marker over the building, shows in the inspector and tooltip, and is published to agents via AGENTS.md — resolving a quest (move it to answered:) fires an emerald shockwave at the building
  • Abilities roster: agent skills (SKILL.md files in vault or ~/.claude/skills/) listed in the agents panel — click to copy an invocation
  • Neural Links: toggle vault backlinks between projects as pulsing violet knowledge arcs — your knowledge graph as city infrastructure
  • Agent fleet presence: every v2 heartbeat session gets its own state-colored orb with an identity tooltip (name/state/action/file); two agents touching the same file surface a deterministic conflict ring and inspector row
  • Daily briefing: one command writes a digest note — status counts, blocked/stale attention list, quest board, git heat

Interactions

  • Single-click a building to select + focus it — the city dims around your selection and connected neighbors stay lit; the camera doesn't move. Double-click opens the note. (This changed in 0.4 — click no longer opens; a one-time hint appears.)
  • Esc or a click on empty ground clears the selection; Move building is a right-click menu item.
  • Right-click menu: Launch agent · Inspect · Move building · Trace impact · Open folder · Open terminal · Copy path · Add quest · Open note · Focus camera.
  • Search and filters narrow the city by title, status, priority, category, path, or stack — filtered-out buildings hide in place (no re-shuffling).
  • Scan lenses (dropdown): status, Needs Attention (triage), Git activity, memory-ready, task-progress ramp, recency heatmap, tech-stack — with an adaptive legend. A ⚠ badge jumps to the attention lens.
  • Saved lens presets: three defaults (Active Work / Needs Attention / Agents) plus "Save current view".
  • EDGES chips toggle the typed project graph: Backlinks · Deps · Blocked · Agents.
  • Project inspector: git signals + recent commits, warnings, dependency sections (Depends on / Used by / Blocked by / Blocks), agent activity, and a last-session digest. City overview (nothing selected) shows district analytics, a fleet summary, an attention list, and a recent-activity feed.
  • Building style (settings): Classic silhouettes or Parametric data-true towers (opt-in, applies live).
  • Snapshot: one click saves a clean cinematic PNG (title card, no HUD) into your vault.
  • Drag handles rearrange category blocks; scroll to zoom, right-drag to pan; keyboard cycles blocked/stale projects + resets camera.

Neural Core & Data Arteries

  • Central geodesic wireframe sphere with RGB chromatic split and rotating rings
  • Data Arteries — animated tube geometry flowing from core to buildings on file changes
  • City states: IDLE (cyan) / STREAMING (cyan fast) / BULK_UPDATE (gold)

Claude Code Integration

  • Activity Monitor polls .hypernovum/agents/ (v2 per-session snapshots) plus the legacy .hypernovum-status.json for real-time agent status
  • Persistent streaming artery while Claude is actively working on a project
  • Activity indicator overlay shows current project and action
  • Terminal Launcher for launching Claude Code, GPT Codex, Antigravity CLI, or a custom agent command
  • Agent context handoff writes .hypernovum/SETUP.md with project metadata, Git signals, and memory context pointers before launch
  • Heartbeat script (scripts/heartbeat.js) for Claude Code hooks integration

Git & Memory Signals

  • Read-only Git activity layer shows recent commit velocity, branch, working-tree state, stale projects, and merge conflict signals
  • Memory-ready filter finds projects that already have .hypernovum/MEMORY_CONTEXT.md
  • Funding metadata is included for users who want to support the free plugin

HUD

  • HYPERNOVUM neon title with flashing block cursor at top center
  • Adaptive SCAN-MODE legend that re-renders per visual layer (status chips, gradient ramps, live stack roster)
  • Controls hint overlay with all mouse and keyboard shortcuts
  • Save Layout and Snapshot buttons

Platform Support

Platform Terminal Emulators Notes
Windows Windows Terminal, cmd.exe Tries wt first, falls back to cmd
macOS iTerm2, Terminal.app Tries iTerm2 first (if running), falls back to Terminal.app
Linux gnome-terminal, konsole, xfce4-terminal, xterm Tries each in order until one succeeds

All features — Neural Core, Data Arteries, Claude Code integration, context menus — work identically on every platform. The only difference is which terminal emulator opens.

Frontmatter Schema

Projects are detected by frontmatter tag project or field type: project. See SCHEMA.md for the full field reference.

---
tags: [project]
title: My Project
status: active
priority: high
category: web-apps
stack: [TypeScript, React, Vite]
questions:            # optional research quests for AI agents
  - "Which vector DB fits this workload?"
projectDir: C:\Users\me\projects\my-project   # Windows
# projectDir: /Users/me/projects/my-project   # macOS
# projectDir: /home/me/projects/my-project    # Linux
---

AI Integration

Hypernovum has no built-in AI. External AI tools (Claude Code, etc.) read SCHEMA.md to learn the frontmatter format, scan your project directories, and write frontmatter to vault notes. Hypernovum renders the result.

Prepare vault for AI agents (command palette, settings, or the agents panel) writes an AGENTS.md at the vault root containing the frontmatter schema, a live inventory of your projects, and instructions for making agent activity visible in the city — so any CLI agent launched in the vault immediately understands your second brain. Safe to re-run: only the marked Hypernovum section is regenerated; the rest of an existing AGENTS.md is preserved.

The scripts/heartbeat.js script wires into Claude Code hooks to render each agent session as a named, stateful orb. Heartbeat v2 gives every session its own snapshot file (.hypernovum/agents/<sessionId>.json), so any number of agents run concurrently without clobbering each other:

# per ping (pass --id so every ping updates the same orb)
node scripts/heartbeat.js --vault="/path/to/MyVault" --id="$CLAUDE_SESSION_ID" \
  --name="Claude Code" --agent-type=claude --project="my-project" \
  --state=editing --tool=Edit --file=src/x.ts

# on finish
node scripts/heartbeat.js --vault="/path/to/MyVault" --id="$CLAUDE_SESSION_ID" --stop

The orb is colored by --state (working / waiting / blocked / complete / stale), carries an identity tooltip, and — when two sessions send overlapping --file values on one project — surfaces a deterministic conflict in the city. The --vault flag must point to your actual vault folder regardless of platform. See the AGENTS.md / per-project .hypernovum/SETUP.md that Hypernovum generates for ready-to-paste hook JSON.

Legacy: the older single-file .hypernovum-status.json is still read for one release, so existing hooks keep working (as an anonymous agent). v2 adds identity, lifecycle state, and conflict detection.

Development

npm install
npm run dev    # watch mode
npm run build  # production build

Built with Three.js, Zustand, and the Obsidian Plugin API.

Hypernovum Pro

This Obsidian plugin remains free and open source. For a standalone desktop experience beyond Obsidian — featuring full Engram Persistent Agent Memory, AI agent management, MCP server integration, Tandem Terminal, broader project scanning, and more — check out Hypernovum Pro.

Hypernovum Pro

Hypernovum Pro Dashboard

License

AGPL-3.0 — Free to use, modify, and distribute. Any modified version that is deployed must also be open-sourced under AGPL-3.0.


Keywords: agent ops, second brain, 3D IDE, Three.js IDE, Obsidian plugin, AI agents, Claude Code, agentic coding, code city, knowledge graph visualization, project dashboard, PKM, developer tools

About

Agent Ops for your second brain - a 3D IDE for Obsidian built on Three.js. Visualize your vault as a code city, dispatch AI coding agents (Claude Code, Codex, Antigravity), post research quests, and watch your agent fleet work in real time.

Topics

Resources

License

Stars

86 stars

Watchers

4 watching

Forks

Packages

 
 
 

Contributors