Full creator runtime
One runtime for the whole creator stack.
Developer · Designer · Filmmaker · Photographer · Illustrator · Podcaster · Musician · Writer · Founder · Multi-vertical.
Ten verticals, one Creator Identity Hub, one MCP runtime.
Your full creator state.
Name, tagline, live status, collaborators, inner ring, banner — the headline of every visit, all on one card.
@greymoth
building · live
Live studio wall.
Work-in-progress visible to your circle. Sketches, sticky notes, half-finished thoughts — your studio wall, public.
Credit, built-in.
Every project and every todo can name who you build with. Credit is structural, not an afterthought.
Guild rooms with AI artifacts.
Discord-style spaces with AI-generated business plans, marketing copy, standup summary — always at hand.
standup-summary
business-plan
requirements.md
−92% tokens on tool calls.
AI Identity avatars + token-aware agent infra. HOT layer, sub-agent context hand-off, full GUI ↔ MCP sync.
> todo_smart_next()
→ 148 tok · -92%
−92% tokens.
(vs naive 1,800-tok context dump)
The MCP server your AI tokens actually deserve.
For developers, designers, filmmakers, photographers, illustrators, podcasters, musicians, writers, founders, and multi-vertical creators.
High craft · high autonomy · real workflow integration
0+MCP tools
~0 tokHOT primer
~0 toknext-action call
0 msmedian latency
0LLM-cost classify
WhyToken waste is an
Token waste is an
infrastructure problem.
Most task systems treat the AI as a display consumer. We built Todo Hub as a context runtime — filter-first, cost-aware from the ground up.
Context overflow
−
Full task list dumped into every prompt. 1200+ tokens before your actual question.
+
~80 tok primer via HOT layer. Lazy tools load only when needed.
Zero-cost classification
−
Calling an LLM to tag "marketing" or "short-term" on each insert. Cost adds up.
+
Rule-based auto-classification on insert. 0 LLM API calls. Instant.
Sub-agent context hand-off
−
Parent agent re-explains the full spec in every child prompt. 600+ tok repeated.
+
delegate_brief loads full spec from DB in child context. Parent sends 30 tok.
HowThree design decisions
Three design decisions
that change everything.
Architecture
HOT / LAZY two-tier
HOT layer: ~80 tok primer with system status, active project, and unblocked next task. Heavy tools stay LAZY — loaded only on demand.
context() // 80 tok — HOT → todo_next_unblocked() // +70 tok → delegate_brief(id) // sub-agent only
Classification
Auto-classify on insert
Domain (marketing, infra, design…) and term (short/mid/long) are tagged at write time via rule engine. LLM never touches this path.
todo_add( title="MCP LP", tags="marketing,design" ) // domain=marketing · term=short // 0 LLM calls
Sub-agents
delegate_brief
Parent sends a 30-tok message with just the task id. Child agent calls delegate_brief(id) and gets the full spec from DB — spec lives once, referenced many times.
// parent (30 tok) delegate_brief(id="ae12e45a") // child receives full spec // files · constraints · output fmt
Features
Everything in one runtime.
Token saving is the headline. But the full feature set covers the entire AI task lifecycle — from insert to sub-agent hand-off to GUI reflection.
0%tokens saved per call
0.0Ktokens saved / day
0agent depth supported
smart filter
Domain & term filtering
todo_smart_next(domain, term) returns only what the current agent needs. 9 domains × 3 term buckets = precise routing.
GUI sync
Bidirectional GUI integration
MCP-added TODOs appear in the PRESENSE web app overview, todo page, and public profile rail with no extra sync step.
navigation
Project ↔ TODO links
Clickable tags in the GUI navigate between projects and their associated tasks. Context stays intact across surfaces.
context
Multi-project scope
switch_project(name) changes runtime scope mid-session. One MCP instance, many projects, zero re-initialization.
persistence
Done-summary retention
todo_done(id, summary="...") prepends your implementation note to the description. Future agents inherit the IP.
sync
Backfill & cross-DB ownership
MCP-added rows auto-sync owner_id to the GUI. No orphaned tasks. Public profiles reflect live task state.
The full canvasMCP is one node.
MCP is one node.
The full creator runtime sits next to it.
Tap a chip to jump between the twelve pillars. Each is a different room — different palette, different mockup, different copy.
Identity Hero
Your full creator state.one card, one canvas.
Name, tagline, live status, collaborators, inner ring, banner — the headline of every visit, all on one card. Visitors land on your work, not a feed.
What it solves
A single source of truth for "who you are right now" — identity, status, collabs, ring.
What it doesn't do
It is not a profile-pic + bio template. It does not pretend you are one-dimensional.
@greymothlive
creator · jpShipping PRESENSE · pro musician · AI builder. Currently building token-aware agent infra.
12
projects
47
collab
14
inner
PostIt Board
Live studio wall.work-in-progress, in public.
Sketches, sticky notes, half-finished thoughts. Your studio wall, public. Drag notes around, color them by mood, pin the ones that matter today.
What it solves
A visual surface for everything that is not done yet — and isn't ready to be a "post".
What it doesn't do
It is not a Trello board. It is not a Twitter feed. It is your wall, not your inbox.
todo: ship LP
fix wave bug
demo @ 14h
mix track 03
AI portrait
inner update
Collaborator Slot
Credit, built-in.on every project, every todo.
Every project and every todo can name who you build with. Credit is structural — not a name buried in a description. Their face shows up on your page, and yours on theirs.
What it solves
Co-creator attribution that survives — instead of vanishing in a screenshot.
What it doesn't do
It is not a "tag a friend" sticker. It is not an org-chart. Co-creators are first-class.
identity hub refactor
+2
MCP token guard
+1
LP redesign
+3
INNER Broadcasts
Signal to your 5–15.no feed, no algorithm.
Daily standup-style updates to your trusted circle. Not your followers. Not the public. The 14 people who actually care about what you shipped today.
What it solves
Updates that go to the people who matter — without becoming a public performance.
What it doesn't do
It is not a tweet. It is not a status. It does not get re-shared into the void.
broadcast · INNER 14
today
shipped wave divider refactor · cleared 7 todos · half-baked sketch for postit physics
14 in your circle
SEND
Async curated drops.zero pressure notifications.
Link, song, sketch, AI output — drop it. No "they typed…" indicator. No ping cascade. The recipient sees it when they choose to, in a calm reading state.
What it solves
A maker-friendly DM model. Quiet, async, file-rich. Built for craft, not chat.
What it doesn't do
It is not chat. It is not a notification engine. It does not measure your read-receipts.
async drop
♪track-draft-03.mp3async
↗concept-sketch.pngasync
✦fal.ai/portrait/47async
to@kairon · @nivisgearno-ping
Multi-platform Widgets
Thirteen, one profile.every surface, one identity.
YouTube, GitHub, WakaTime, Spotify, Twitch, SoundCloud, Bandcamp, DEV.to, Zenn, note, HN — all live on one canvas. Status is real, not a screenshot.
What it solves
A single canvas where all your work-surfaces show up live, in your own design.
What it doesn't do
It is not a Linktree. It is not a list of icons. It is a real-time mosaic of your craft.
GitHub
YouTube
Twitch
Spotify
WakaTime
Bandcamp
DEV.to
note
47 Layouts × Textures
Your own design system.not the template everyone picked.
Layouts × textures × fonts × palettes. Cycle until your profile feels like you. Real typographic control, real palette tokens, no template-debt.
What it solves
Structural design control that actually changes how your page looks — not a skin.
What it doesn't do
It is not 5 themes with a different accent color. It is not a color picker hidden behind upgrade.
47 layouts · 12 textures · ∞ combos
Group + AI Artifacts
Guild rooms with AI artifacts.whiteboard meets agent.
Discord-style spaces with AI-generated business plans, marketing copy, standup summary — always at hand. Long-running artifacts your team can read, edit, version.
What it solves
A collaboration space where AI output is a long-lived artifact, not a chat scroll.
What it doesn't do
It is not a chatroom. It is not a Notion doc. It is a working whiteboard with agents.
#guild · greymoth-os3 online
standup-summary.mdAI · today
business-plan-v3.mdAI · 2d
requirements.mdhuman · 1w
marketing-copy.mdAI · draft
MCP Runtime
−92% tokens on tool calls.the runtime your AI deserves.
AI Identity avatars + token-aware agent infra. HOT layer, sub-agent context hand-off, full GUI ↔ MCP sync. Built so your agents stop burning context on boilerplate.
What it solves
A context runtime — filter-first, cost-aware, GUI-synced. Tokens that respect you.
What it doesn't do
It is not an LLM. It is not a chat UI. It is the cheap, fast plumbing in between.
claude · mcp tool call
> todo_smart_next(domain="marketing")
// HOT layer · no LLM classify · pure DB
id: ae12e45a
domain: marketing
tokens_used: 148
naive dump: 1,800 tok−92%
In Practice
One call. Full context.
claude — mcp tool call
> todo_smart_next(domain="marketing")
// → 150 tok. No LLM classification. Pure DB.
id: ae12e45a
title: MCP LP — nivisgear × fal.ai
domain: marketing
term: short
priority: high
tokens_used: 148
> todo_claim(id="ae12e45a")
> _
Reference Implementation
presense.dev ↗PRESENSE is built on this.
Every feature above powers the PRESENSE Creator Identity Hub in production. The MCP runtime drives real AI agent workflows — not a demo.
Verticals
Built for every kind of creator.
Not a dev-only tool. Not a freelance designer board. PRESENSE adapts its Project schema, Journal categories, and AI agent personality to your craft.
Developer
Write code, build products, ship
Ship-log + GitHub + WakaTime in one runtime.
Designer
Visual systems, UI, brand identity
Case studies, design system, freelance brief in one canvas.
Filmmaker
Cinematography, directing, editing
Reel + festival tracker + crew rolodex.
Photographer
Capture light, tell visual stories
Booking calendar, gallery, client delivery.
Illustrator
Characters, worlds, editorial art
Editorial portfolio + commission queue.
Podcaster
Audio, narrative, voice craft
Episode archive, transcript, guest network.
Musician
Compose, produce, perform, record
Catalog, Now Spinning, listening rooms.
Writer
Essays, fiction, scripts, copy
Essay queue + drafts + reading list.
Founder
Build companies, ship products
Roadmap, fundraising, traction in public.
Multi / Other
Combination of all of the above
For builders who refuse to pick one lane.
AI AgentMulti-model, per-tier.
Multi-model, per-tier.
Right brain for the job.
We don't lock you to one provider. The Group AI Agent routes simple summaries to Gemini Flash, planning to Gemini 2.5 Pro, and Opus-grade artistry to Claude 4.7 — based on your tier and the task complexity.
Free
$0/moGemini 2.0 Flash
Smart enough for daily standups. Free forever, no credit card.
Creator
$12/moGemini 2.5 Pro
Gemini 2.0 Flash
Reasoning-grade for project planning + creative drafts.
Studio
$25/moClaude Opus 4.7
Claude Haiku 4.5
Gemini 2.5 Pro
Opus-grade artistry. Multi-model routing per task complexity.
Affinity Matching
Collab matches that actually fit.
Your declared verticals + intent + every project tag, todo tag, and journal keyword feed a cosine-similarity vector. Each write auto-marks your profile dirty; the next match query rebuilds and ranks top-K candidates.
Weights: declared tags 3.0 · vertical 5.0 · intent 4.0 · project tags 1.5 · todo tags 1.0 · journal keywords 0.8
affinity rebuild · 1 user
vertical designer → 5.0
intent project-collab → 4.0
tags [Figma, brand, motion] → 9.0
projects [case study, design system] → 3.0
journal [type, layout, color] → 2.4
// normalized → unit vector
cosine_sim(viewer, candidate) → 0.87
// top-K = 10 returned
Per-Project Journal
Every entry lives inside a project.
Click any project on your overview → Journal tab. Devlog notes, session recaps, behind-the-scenes — all scoped to the project they belong to. No more flat blog feed where everything dilutes everything else.
Project as the spine
Every artefact — todo, journal entry, banner, collaborator, vertical — hangs off a single Project row. The hub, not the feed.
Journal tab per project
Projects → click → Journal. Only the entries linked to that project. AI agents reading via devlog.list(projectId) stay tightly scoped.
MCP-controllable
Every field above is mutable from MCP: project.update, project.add_link, project.set_vertical, project.banner_set.
If you're a photographer…
Your booking calendar, client gallery, and delivery proofs live in one Project. Each session gets its own Journal entry — call sheet, license, payment, all linked.
Booking calendarClient galleryDelivery proofsLicense logs
If you're a filmmaker…
Reel + festival submission tracker + crew rolodex. Every short, every spec deck, every pitch lives as a Project with its own journal, custom links, and collaborator avatars.
ReelFestival trackerCrewPitch decks
If you're a designer…
Case studies as live Projects. Design system as a separate Project. Freelance briefs as journal entries. Every shot reads like a portfolio site without exporting one.
Case studiesDesign systemFreelance briefsMood boards
If you're a podcaster…
Episode archive as a Project. Each episode = a journal entry with transcript, guest, runtime, RSS link. Guests stay in the collaborator graph.
Episode archiveTranscriptsGuest graphRSS sync
Pick a path. Start now.
Free forever, no credit card. Upgrade when you outgrow it.
Get early access.
PRESENSE · MCP RUNTIMEMCP server is live for PRESENSE. Standalone release in progress.