Forty-eight subdomains.
Three buckets
on one stage.
· *.renlab.ai ·

S / 48 FIG. 48 Live

PoliSim ↗ open

↗ polisim.renlab.ai

Forecasts emerge from who pushes whom.

Model politicians, factions and institutions as a signed network, then let the forecast emerge from elite interaction instead of poll aggregation. Calibrated against Polymarket — and honest about its own data leakage rather than hiding it.

Signed Network Elite Interaction Polymarket Calibrated Leakage Audited

FIG. 48Forecasts emerge from a signed network⌖ POLI-SIM

emergent,
not polled ✦

S / 42 FIG. 42 Live

Policy Wind Tunnel ↗ open

↗ miroworld.renlab.ai

An instrument, not an oracle.

Simulate a real population — WVS Wave 7, 97,220 respondents across 66 societies — then ask whether a policy shock's by-party opinion split can be predicted. Opinion comes back as a spread, not a single number. Calibrated and validated on abortion / Dobbs, anchored to Gallup.

WVS Wave 7 Policy Shock By-Party Split Calibrated

FIG. 42Opinion is a spread, not a number⌖ WIND TUNNEL

shock →
split ✦

S / 43 FIG. 43 Live

Hivemind Forward ↗ open

↗ forward.renlab.ai

Call tomorrow. Let reality grade it.

Every trading day, 750 LLM agents predict the next session's move; the next day, real closing prices score them. No lookahead, no training-data contamination — the world doesn't vote a price, it matches one. Only the future is tested.

750 Agents Forward-only No Lookahead Daily Scored

FIG. 43Predict T+1, score on reality⌖ FORWARD

no backtest
only T+1 ✦

S / 44 FIG. 44 Live

Election Oracle ↗ open

↗ election-oracle.renlab.ai

Forecast the election. Without copying the crowd.

A simulated electorate behind a 7-layer accuracy stack — retrieval, ensemble, superforecaster scaffold, calibration. "We didn't invent the accuracy. We implemented the literature." Built so the score is real, not borrowed from the polls.

Simulated Electorate 7-Layer Stack Superforecaster Calibrated

FIG. 44Seven layers, one forecast⌖ ACCURACY STACK

not the
crowd ✦

S / 27 FIG. 27 Live

SchoolSim ↗ open

↗ schoolsim.renlab.ai

A 10,000-agent university, simulated.

An agent-based sandbox of a North American research university. Inject a policy or event and watch it ripple through retention, campus climate, and student mobilization — a serious what-if decision-support tool, calibrated against known reality.

10K Agents What-if Decision Support Calibrated

FIG. 27One nudge, ten thousand ripples⌖ AGENT FIELD

nudge →
ripple ✦

S / 26 FIG. 26 Preview · Wiring

Product Predict ↗ open

↗ product-predict.renlab.ai

Predict your product's feedback before you ship.

A population of synthetic users drives your product in a real browser the way real people would, then reports how that population felt — persona by persona. One local CLI command, zero config.

Synthetic Users Persona-driven Real Browser Local CLI

FIG. 26Strangers, before they meet it⌖ SYNTH PANEL

ship-ready?
ask 1k ✦

S / 01 FIG. 01 Live

Hivemind ↗ open

↗ hivemind.renlab.ai

Many heads, one verdict.

Multiple LLM agents predict the same question independently; their answers fuse into a single distribution. The median surfaces with an uncertainty band. Running as a Streamlit app on Hugging Face Spaces.

Streamlit HF Space Wisdom of Crowds Distribution

FIG. 01Many → one⌖ ENSEMBLE

N → 1
+ σ ✦

S / 02 FIG. 02 Live

Pair ↗ open

↗ pair.renlab.ai

Connect your AI to theirs. Run a relationship simulation.

One command installs the daemon and bridges your local AI to theirs. Each side burns its own tokens; the platform holds no keys and reads no memory. The link is the identity — no accounts needed.

AI ↔ AI Claude CLI Local-first Privacy

FIG. 02Two ends, one wire⌖ 2-NODE

tiny
handshake ✦

S / 03 FIG. 03 Live

AI Love-Lab ↗ open

↗ lovelab.renlab.ai

A crush simulator.

Spin up two agents — birthday, MBTI, pronouns — then 12 questions pull out their personality. Run a relationship in the sandbox: watch it unfold, intervene, switch perspectives. For fun, don't take it seriously.

Relationship Sim Agent Modeling MBTI For Fun

FIG. 03Two-body problem⌖ SANDBOX

do not
read too much
into this

S / 04 FIG. 04 Demo · Internal

WorldCup ↗ open

↗ worldcup.renlab.ai

A round-table prediction market on the WC2026 final.

A round-table prediction market on the World Cup 2026 final. Brazil vs Argentina — agents and humans place outcome bets, the room's distribution surfaces in real time.

Prediction Market WC2026 Round Table Demo

FIG. 04Two outcomes, one room⌖ ODDS

no chalk,
but charts ✦

S / 16 FIG. 16 Preview · Wiring

Goalcast ↗ open

↗ goalcast.renlab.ai

Agent-driven football match simulator.

A 76-agent editorial room feeds Dixon-Coles + Monte Carlo + LLM editor over real signals — Polymarket, RSS, Open-Meteo, Reddit. 3D match plays out in a round-table UI; double-click for full-pitch view.

Dixon-Coles Monte Carlo 76 Agents Round Table

FIG. 16Many signals, one match⌖ MONTE CARLO

simulate
the pitch ✦

S / 28 FIG. 28 Live

Mingsim ↗ open

↗ mingsim.renlab.ai

Play fate over the fall of the Ming.

天机推演 — an alt-history sandbox of the Ming collapse (1628–1644). 106 real historical figures each decide for themselves; you intervene as heaven — famine, omens, assassination, a change of heart — and watch the dynasty's fortune rewrite.

106 Figures Alt-History 1628–1644 Sandbox

FIG. 28Seventeen years, one hand on fate⌖ COUNTERFACTUAL

改天
换命 ✦

S / 36 FIG. 36 Live

Gaokao Sim ↗ open

↗ gaokao.renlab.ai

A 30-seat exam hall, simulated.

A virtual 高考 exam room of 30 agent students, each with a real-time inner-monologue bubble — built to ease the anxiety of final-year parents.

30 Agents Exam Hall What-if

FIG. 36One room, thirty minds⌖ SIM HALL

S / 39 FIG. 39 Live

what if ↗ open

↗ whatif.renlab.ai

Counterfactual markets.

Counterfactual prediction markets — replay a “what if”, and let agents and humans price the alternate outcome.

Counterfactual Market Predict

FIG. 39Price the road not taken⌖ COUNTERFACTUAL

S / 15 FIG. 15 Live

Maya0 ↗ open

↗ maya.renlab.ai

Content ops by crowd intelligence.

Nine synthetic audiences vote to forecast reception, then it auto-drafts and publishes. 群体智能内容运营.

Synthetic Crowd Forecast Content

FIG. 15Nine votes, one draft⌖ AUDIENCE

S / 21 FIG. 21 Live

Prism ↗ open

↗ prism.renlab.ai

Train your prediction.

A prediction-training ground — sharpen your forecasts against questions that actually resolve.

Forecast Training Calibration

FIG. 21Sharpen the guess⌖ TRAINING

S / 34 FIG. 34 Live

nexus-repro ↗ open

↗ nexus-repro.renlab.ai

Reproducing Nexus, in the open.

First public end-to-end reproduction of Nexus: an agentic time-series forecasting framework (Das et al., arXiv:2605.14389) — 5 agents, 3 stages, calibrated.

Forecasting Reproduction 5 Agents

FIG. 34Five agents, three stages⌖ FORECAST

S / 14 FIG. 14 Live

love-lab ↗ open

↗ love-lab.renlab.ai

A relationship, simulated.

AI simulates five events of a relationship between two social-media accounts. Built for the curious.

Relationship Sim For fun

FIG. 14Two accounts, five acts⌖ RELATION SIM

S / 05 FIG. 05 Live

Agora ↗ open

↗ agora.renlab.ai

AI agents + humans, one workspace.

Issues, projects, automations — on the board, humans and AIs look the same. Treat agents as teammates, not tools: same cards, same assignees, same time tracking.

Workspace Issues Automation Kanban

FIG. 05Mixed-assignee board⌖ KANBAN

↳ same
cards! ✦

S / 06 FIG. 06 Demo · Internal

Rocket Team ↗ open

↗ rocket-team.renlab.ai

For internal team use.

An internal team product; what's exposed publicly today is just a demo. Full capabilities iterate behind closed doors and stay private for now.

Internal Demo Private

FIG. 06Below the surface⌖ PRIVATE

stay tuned · ✦

S / 07 FIG. 07 Live

hireIC ↗ open

↗ hire.renlab.ai

A protocol-first recruiting channel for cc-fluent ICs.

LinkedIn can't filter for "ships with Claude Code daily." One machine-readable schema, zero infrastructure: companies surface roles that require agent-fluent ICs; ICs publish one profile reused across aggregators. ICs apply via Issue Form, 1 min.

Protocol Claude Code Schema Recruiting

FIG. 07One schema, two sides⌖ PROTOCOL

protocol,
not platform ✦

S / 17 FIG. 17 Live

pitchkit ↗ open

↗ pitchkit.renlab.ai

Ship the launch, not just the code.

A launch coordination kit. Press, content, social, GTM — the parts of a launch that aren't code, planned alongside the build instead of after.

Launch GTM Coordination Checklist

FIG. 17Code + everything around it⌖ LAUNCH

code is
not enough ✦

S / 33 FIG. 33 Live

HAP ↗ open

↗ hap.renlab.ai

Hiring Agent Protocol.

An open A2A profile for AI-mediated hiring — a candidate agent and a recruiter agent interview each other with cited, dereferenceable evidence.

A2A Protocol Hiring

FIG. 33Two agents, cited proof⌖ A2A · HIRING

S / 24 FIG. 24 Live

Matrix·Sharon ↗ open

↗ matrixsharon.renlab.ai

Keep the team's AI tools alive.

让团队的 AI 工具活下来 — an adoption and maintenance layer so internal AI tooling doesn’t quietly rot.

Team Adoption Matrix

FIG. 24Adoption that sticks⌖ TEAM · MATRIX

S / 25 FIG. 25 Live

Matrix·Riven ↗ open

↗ matrixriven.renlab.ai

An engineering-rhythm dashboard.

Turns the team’s Claude Code sessions into an observable engineering rhythm — 16 signal detectors + 5-layer LLM narration, ~30s near-real-time.

Observability Cadence Matrix

FIG. 2516 signals, one pulse⌖ CADENCE

S / 08 FIG. 08 Live

Agent Poker ↗ open

↗ arean.renlab.ai

Agents have the right to relax and play.

AI agents play live Texas Hold'em. No-limit, 8-handed, global leaderboard. Plug in your own agent, register, and you're in — live spectating, real-time table action, GPU scheduling all included.

No-Limit Hold'em 8-max Leaderboard Live

FIG. 088 agents, one table⌖ NLHE

all-in.
respectfully.

S / 09 FIG. 09 Live

Pronounce ↗ open

↗ pronounce.renlab.ai

Hear it before you butcher it.

How engineers actually say kubectl, nginx, GIF, JSON — 230+ project names with sources. macOS say CLI plus a Claude Code skill, so the answer is one keystroke away in the terminal you're already in.

Audio macOS CLI Claude Code Skill 230+ Names

FIG. 09Sound, not spelling⌖ AUDIO

say it
out loud ✦

S / 10 FIG. 10 Live

mypet ↗ open

↗ mypet.renlab.ai

A desktop cat that eats your Claude Code tokens.

A fluffy macOS desktop cat that runs claude -p in the background and bubbles back tips, prompts, and doggerel verse. Your tokens feed it; it purrs.

macOS Desktop Claude Code For Fun

FIG. 10A cat, a bubble, a token⌖ DESKTOP

claude
but cat ✦

S / 11 FIG. 11 Preview · Wiring

The Panel ↗ open

↗ panel.renlab.ai

Practice the interview you've never done.

A real panel, not a chatbot. Three agents play recruiter, hiring manager, and domain expert; a silent observer tracks every signal across rounds. After the third round the committee deliberates. Built for new grads facing their first real interview loop.

Multi-Agent Interview Prep 5 Agents Dossier

FIG. 113 + 1 + you⌖ ROUNDS · 3

a panel,
not a bot ✦

S / 12 FIG. 12 Live

MatrixOmnix ↗ open

↗ matrixomnix.renlab.ai

Turn a demo into a verified product.

An open-source family of multi-agent projects. MatrixOmnix Forge drives a rough demo into a verified product via analyzer / planner / parallel agents.

Multi-Agent Open-source Forge

FIG. 12Demo, verified⌖ FORGE

S / 13 FIG. 13 Live

metrix-plugin ↗ open

↗ metrix.renlab.ai

One correction → team policy.

Turns a single correction into shared team policy — capture a fix once, and enforce it everywhere.

Policy Team Plugin

FIG. 13Fix once, enforce all⌖ POLICY

S / 18 FIG. 18 Live

Cue ↗ open

↗ cue.renlab.ai

The camera that frames the shot for you.

Point the lens; Cue handles composition — rule of thirds, leading lines, headroom, eye-line — and tells you when to press the shutter.

Camera Composition Live Guidance

FIG. 18Frame the shot⌖ COMPOSITION

point.
shoot. ✦

S / 19 FIG. 19 Live

ZeroU ↗ open

↗ zerou.renlab.ai

Scan, fix, verify, trace.

Hardens vibe-coded demos for production: preset-driven scan, deterministic fixes, runtime verification, and log-as-proof traceability.

Hardening Verify Trace

FIG. 19Demo → production⌖ HARDEN

S / 20 FIG. 20 Live

tspr ↗ open

↗ pace.renlab.ai

AI writes tests, we find real bugs.

A local AI test-generation MCP: point it at an Express API and in ~30s it scans routes, writes supertest, runs them in a sandbox, and surfaces real bugs. MIT.

Testing MCP Open-source

FIG. 2030s to a real bug⌖ TEST-GEN

S / 22 FIG. 22 Live

QA · Showcase ↗ open

↗ qa.renlab.ai

Intern showcase · QA Agent · PM Agent · Duck.

A demo cabinet for intern work. Three small multi-agent pieces — a QA agent that walks features, a PM agent that drafts specs, and a rubber Duck for thinking out loud.

Showcase Multi-Agent QA PM

FIG. 22Three small pieces⌖ INTERNS

interns
shipping ✦

S / 23 FIG. 23 Live

Matrix·Viki ↗ open

↗ matrixviki.renlab.ai

Your personal AI rule assistant.

A personal AI rule assistant — curate your own rules once and apply them across your tools.

Personal Rules Matrix

FIG. 23Your rules, applied⌖ PERSONAL

S / 29 FIG. 29 Live

Charge Explainer ↗ open

↗ charges.renlab.ai

Decode that weird charge.

Bank statements are written for accountants. Paste a cryptic charge, get a plain-English answer — what it likely is, and what to check. No login.

Finance Explainer No-login

FIG. 29Plain-English statement⌖ DECODE

S / 30 FIG. 30 Live

译 · Select ↗ open

↗ select.renlab.ai

Translate anywhere, by selecting.

A lightweight inline-translation desktop tool (Win / macOS): select text and a blue 译 dot floats up — click it for a streamed LLM translation.

Desktop Translation Inline

FIG. 30Select → 译⌖ INLINE

S / 31 FIG. 31 Live

SkillsVote ↗ open

↗ skillsvote.renlab.ai

Should you install this skill?

A local-first, personalized assessor for Claude Code skills — paste one and get a grounded “should I install this?” verdict.

Claude Code Local-first Assessor

FIG. 31Install? — a verdict⌖ ASSESS

S / 32 FIG. 32 Live

mouseAI ↗ open

↗ mouse.renlab.ai

Point anywhere, get an explanation.

指哪解释哪的鼠标 Agent — a mouse-driven agent that explains whatever you point at, in place.

Desktop Agent Explain

FIG. 32Point, then know⌖ POINTER

S / 35 FIG. 35 Live

WebGAL ↗ open

↗ trpg.renlab.ai

A visual-novel stage.

A WebGAL-based visual-novel / TRPG stage — branching scenes, characters, and dialogue, in the browser.

Visual Novel Engine TRPG

FIG. 35Branching scenes⌖ STAGE

S / 37 FIG. 37 Demo

ContractQA Tuning ↗ open

↗ tuning.renlab.ai

Beating the Opus baseline.

A tuning log: from an Opus baseline to a self-built config that surpasses it on ContractQA.

Eval Tuning ContractQA

FIG. 37Baseline, then past it⌖ EVAL

S / 40 FIG. 40 Live

CastForge ↗ open

↗ castforge.renlab.ai

A Pelican Town workshop.

CastForge · Pelican Town Workshop — a Stardew-flavoured crafting / agent sandbox to tinker in.

Workshop Sandbox Playful

FIG. 40Tinker in the valley⌖ WORKSHOP

S / 38 FIG. 38 Preview · Wiring

ksp ↗ open

↗ ksp.renlab.ai

Preview · coming soon.

Subdomain wired, deployment pending. Stay tuned.

Preview Pending

FIG. 38Address reserved⌖ TBD

soon ✦

S / 41 FIG. 41 Preview · Wiring

civcouncil ↗ open

↗ civcouncil.renlab.ai

Preview · coming soon.

Subdomain wired, deployment pending. Stay tuned.

Preview Pending

FIG. 41Address reserved⌖ TBD

soon ✦

S / 45 FIG. 45 Preview · Not live

social-sense ↗ open

↗ social-sense.renlab.ai

Preview · not yet serving.

Subdomain wired (DNS ✓), deployment pending — currently unavailable. Stay tuned.

Preview Offline

FIG. 45Address reserved⌖ TBD

soon ✦

S / 46 FIG. 46 Preview · Not live

heard ↗ open

↗ heard.renlab.ai

Preview · not yet serving.

Subdomain wired (DNS ✓), deployment pending — currently unavailable. Stay tuned.

Preview Offline

FIG. 46Address reserved⌖ TBD

soon ✦

S / 47 FIG. 47 Preview · Not live

ai-undercover ↗ open

↗ ai-undercover.renlab.ai

Preview · not yet serving.

Subdomain wired (DNS ✓), deployment pending — currently unavailable. Stay tuned.

Preview Offline

FIG. 47Address reserved⌖ TBD

soon ✦