Live Experiments

DuckDBAnthropicOpenAIAsyncLocalStorageCost Observability

Token FinOps Proxy

A thin proxy layer in front of @diabolicallabs/llm-client (Anthropic + OpenAI) tags every call with (agent, persona, model, scope, route, intent) via AsyncLocalStorage and persists token and cost to an embedded DuckDB store, surfacing cost by capability, by model-tier shift, and by incident class instead of a per-provider spend total.

EXP_045

p5.jsSilero VADonnxruntime-webOn-Device MLGenerative ArtCreative Tech

On-Device Presence Sensing

A local, open-weight `Silero VAD` model runs entirely in your browser via `onnxruntime-web`, reading your voice's presence and rhythm frame by frame with no cloud round-trip. This demonstrates the same physical-digital feedback loop Diana builds for experiential installations, driving a `p5.js` generative canvas in her brand palette on genuine local inference rather than an API call.

EXP_044

agent-opscost-optimisationmodel-routinglocal-inferencebenchmark

Local Edge-Model Agent Router (MiniCPM5-2B)

A self-hosted 2-billion-parameter MiniCPM5-2B instance triages 16 real tasks, deciding per task whether it can answer directly or must escalate to the fleet's cloud model, with every task's real cloud cost and latency measured regardless of the router's decision — a genuine before/after comparison, not an estimate.

EXP_043

MCPAgent OpsPolicy EngineAccess ControlZod

MCP Tool Gateway

This experiment gates every MCP tool call behind a policy engine that checks agent, server, and tool identity before proxying to a real in-repo demo server, denying by default whenever the ruleset is corrupted or a real fleet server has no live credentials in this public deployment.

EXP_036

agent-opscontext-engineeringdeveloper-toolsinteractive

Modular-Context-File Compiler

Pick a session context and watch a five-layer stack of Diana's own agent context files compile into one resolved view, with every line tagged by the layer it came from and every override traced back to the earlier layer it replaced.

2026.09.12

EXP_035

agent-opspaymentsspend-capallowlistinteractive

Agent-to-Agent Micro-Payment Simulator

Two mock agent wallets, Atlas and Scout, transact against their own spend caps and merchant allowlists while a scripted eight-transaction run shows the ledger allow one payment, deny another for an unlisted merchant, and deny a third for an exhausted cap, each with its running balance.

2026.09.08

EXP_042

Canvas 2DOpenAIZodGenerative ArtReal-Time Adaptation

Adaptive Installation Controller

An LLM planner reads a simulated real-time sensor/audience stream — footfall, dwell time, ambient noise, time of day — and drives a Canvas 2D installation scene's lighting, colour, motion, and pacing on a bounded cadence, toggleable against a fixed scripted cue stack that ignores the sensor reading entirely.

2026.09.07

EXP_041

agent-opsdata-residencylocal-inference@diabolicallabs/llm-clientmodel-routing

Hybrid Local/Cloud Data-Sensitivity Router

Types a question and a rule-based classifier decides, by data sensitivity alone rather than cost or latency, whether a local model or a cloud model answers it, then shows the full decision trace: what was flagged, why, and which model actually responded.

2026.09.07

EXP_040

MCPOpenAIAgent OpsZodServer-Sent Events

Mock MCP Retail-Inventory Connector

This experiment demonstrates a live agent tool-calling loop against a mock, Zod-validated MCP-shaped inventory server — search, stock check, and reservation tools — that reacts on a simulated in-store display streamed over Server-Sent Events. It reconstructs the MCP tool-surface shape, not an integration with any real retailer.

2026.09.07

EXP_034

agent-opssecurityzero-trustidentityinteractive

Zero-Trust Agent Gateway

Watch three mock agents present simulated identities to a gateway that checks each one, routes routine actions straight through, and blocks sensitive actions until you personally approve or deny them — while an agent with an expired or unregistered identity is rejected before its requested action is even read.

2026.09.04

EXP_039

Canvas 2DOpenAIAgent OpsZodHardware Control

MHS Mock Hardware Controller

This experiment demonstrates a full protocol cycle — device discovery, capability negotiation, and structured read/write commands — sent to a simulated kinetic installation of two motors and a light fixture, rendered live on a Canvas 2D viewport. It models the shape of Anthropic's Model Hardware Standard research-preview announcement, not an integration with the real MHS SDK.

2026.09.03

EXP_033

agent-opsmemoryknowledge-graphcogneeinteractive

Shared Agent Memory Graph

Two mock agents, Curator and Auditor, read and extend one seeded brand-asset knowledge graph of guidelines, decisions, campaigns, and channels. Trigger their scripted queries and watch two of them render a live graph traversal beside a flat-similarity search over the identical text, so the gap between structured relationships and flat search is something you see, not something the copy just asserts.

2026.08.30

EXP_032

agent-opsresiliencereliability@diabolicallabs/llm-client

Fallback Model Router

A visitor triggers three model calls in sequence: a healthy primary response, a simulated primary failure that reroutes live to a lower-tier fallback model, and a simulated total outage that fails fast with a clear error — every result labelled with the model that actually served it.

2026.08.26

EXP_038

agent-opscost-optimisationmodel-routingbenchmarklocal-inference

Local Qwen3.8-27B Agent-Routing Benchmark

22 real historical fleet-dispatch decisions — redacted briefs paired with their actual `delegate:` ground truth — run through a self-hosted Qwen3.8-27B on Ollama and through the fleet's default cloud model, scoring which specialist each one predicts and measuring the real accuracy, latency, and cost trade-off of routing that one decision locally instead of to the cloud.

2026.08.25

EXP_037

agent-opssecuritydockersandboxingmulti-agent

Multi-Agent Coordination-Failure Simulator

Two independent worker agents given an identical, under-constrained task in a shared Docker workspace, with no channel to talk to each other, were logged overwriting one another's already-completed output — real independent adversarial convergence when isolation boundaries fail. A second, isolated-container condition shows the same interference structurally cannot occur once each agent gets its own private workspace.

2026.08.25

EXP_031

agent-opsgovernancetrust-tiersinteractive

Reversibility-Tiered Approval Gate

Watch a scripted mock agent work through nine tool calls — reading files, editing config, deleting data, publishing externally — with each one classified by reversibility tier and gated as auto-run or approval-required, and personally approve or deny the ones the gate blocks on.

2026.08.21

EXP_030

agent-opstool-callingbenchmark@diabolicallabs/llm-client

Programmatic-vs-JSON Tool-Calling Benchmark

Six data-lookup tasks over a fixed mock order dataset run twice each against two models — once through native JSON tool-calling, once by having the model write code that calls typed function stubs directly in a sandbox — reproducing arXiv 2608.06370's finding that programmatic tool calling needs far fewer LLM round trips per task than the JSON tool-calling pattern most agent frameworks default to.

2026.08.17

EXP_029

agent-opscost-optimisationmodel-routing@diabolicallabs/llm-client

Effort-Level A/B Harness

Runs a fixed six-task benchmark at four reasoning-effort levels against Claude Opus 5, recording real cost, latency, and a rubric-scored quality result for every task-effort pairing, then charts the trade-off curve.

2026.08.12

EXP_028

securityagent-opsgovernanceinterceptionpolicy

Local Security-Policy Interception Middleware

Watch a mock agent propose tool calls — reads, writes, deletes, publishes, credential access — and this experiment intercepts each one in real time, scores it against an editable local policy schema, and blocks anything at or above your threshold until you approve or deny it.

2026.08.05

EXP_027

agent-opscost-optimisationmodel-routingbenchmark

Model-Routing Cost/Latency Benchmark

A fixed task set of five experiential-design briefs runs against three models — Claude Opus 5, GPT-5.6 Sol, and Gemini 3.6 Flash — and a deterministic rubric scores each response, demonstrating the real cost, latency, and quality trade-off behind routing a task to a bigger model instead of a cheaper one.

2026.08.03

EXP_026

Canvas 2DWeb AudioGenerative ArtAmbient

Bloom Field

A Canvas 2D generative garden of procedural blooms that grows on its own — no login, no product, nothing agentic. Enable your microphone and sound layers a growth pulse, hue shift, and sway on top of the ambient rhythm; decline, and the garden keeps breathing regardless.

2026.08.02

EXP_025

securityagent-opsdesign-biasstatic-analysis

Skill Prompt Auditor

Paste an agent skill's prompt text — a SKILL.md body or system prompt — and get a flagged report of hardcoded design biases: colour palettes, layout defaults, typography directives, tone constraints, and output-format lock-ins, checked entirely in your browser before you ever run the skill.

2026.07.31

EXP_024

agent-opscost-optimisationtoken-reuseprompt-caching@diabolicallabs/llm-pricing

Token-Reuse Cost Visualizer

Upload a token-log export or load the bundled sanitized sample, and this experiment charts reuse ratio over time, reused-versus-fresh token composition, and an estimated cost split — entirely in your browser.

2026.07.28

EXP_023

securityagent-opsgovernancetrust-tier

Trust-Tier Compliance Linter

Paste an agent's declared trust tier and its described execution or publish behaviour, and this experiment scores whether that behaviour actually matches Diana's own Autonomous / Always-Review / Conditional rules, not a generic industry checklist.

2026.07.25

EXP_022

GEOagent-ops@diabolicallabs/llm-clientschema.org

MCP Capability-Discovery Visualizer

Paste or upload an MCP configuration — a JSON list of servers and the tools each one declares — and this experiment shows a server-to-tools capability map, labelling any unreadable server as unavailable instead of breaking the whole view.

LLM Citation-Integrity Metadata Generator

Paste content or a URL and this experiment extracts candidate citable facts, then recommends schema.org markup and source-attribution phrasing for each one — it shows how to prepare content for AI citation before a crawl ever happens.

2026.07.19

EXP_020

agent-opsmulti-agentorchestration@diabolicallabs/llm-client

Declarative Multi-Agent Router

Submit a task and watch a real LLM call route it against a declarative task-shape-to-capability table, dispatch that route's real sub-agent calls in sequence or parallel, and render a live trace of the decision and every sub-agent's actual output — not a diagram of how routing could work, but routing that runs.

2026.07.16

EXP_019

securityagent-opsOWASPstatic-analysis

ASI-Checklist Linter

Paste or upload an agent's configuration — tool definitions, MCP servers, hooks, memory writes — and get a ten-category scorecard against the OWASP Top 10 for Agentic Applications, with a labelled finding for every risk pattern the static analysis actually finds.

2026.07.15

EXP_018

agent-opscost-optimisationmodel-routing@diabolicallabs/llm-client

Reasoning Budget Router

Submit a task and watch a heuristic classifier score its complexity, route it to the cheapest model tier that can handle it, then run the real call and chart the cost and latency you'd have paid at each tier — the fleet's own model-tiering instinct made explicit and measurable.

2026.07.13

EXP_017

GEOagent-opsmulti-provider@diabolicallabs/llm-client

LLM Citation Auditor

Enter a brand name and audit which sources Perplexity and ChatGPT actually cite when asked about it — a footprint map of publications, review sites, and URLs each model draws on, built entirely on live queries with no cached results.

2026.07.11

EXP_016

OpenAIGeminiClaudeSSEGEORecharts

AI Impression Simulator

Type a brand name and watch ChatGPT, Gemini, and Claude answer the same three questions a real buyer would ask — discovery, consideration, comparison — then see each model's visibility score, sentiment, and consistency compared side by side.

2026.07.10

EXP_015

LLMCanvas 2DZodNarrative DesignBrand WorldPrompt Injection Defense

Narrative Micro-World

A small navigable brand world — three scenes, dialogue and environment generated live by an LLM in response to your choices, held to an authored style/lore block so it never drifts off-brand.

2026.07.09

EXP_014

agent-opse-commerceevaluationGEO

Machine-Buyer Storefront A/B

A scripted LLM shopping agent buys from two mock storefronts with identical inventory and prices — one built with countdown timers and obscured pricing, one built for machine legibility — and the harness measures which storefront it actually completes checkout on.

2026.07.04

EXP_013

Canvas 2DRule-Based AgentDeterministic SimulationSpatial Events

Spatial Telemetry Playground

A simulated gallery floor where twelve-plus autonomous synthetic visitors move under distinct `wander`, `cluster`, and `queue` behaviours, streaming structured spatial events — zone entry/exit, dwell, crowding — to a rule-based agent loop that adjusts lighting, display content, and routing in real time. Zero LLM cost by default; deterministic seeded runs make the whole loop testable.

2026.06.28

EXP_012

securityagent-opstoolinggovernance

Agent Permission Linter

Paste a system prompt and tool definitions. Get a flagged list of privilege escalation risks and missing scope constraints — before your agent reaches production.

2026.06.22

EXP_011

p5.jsWebRTCGeminiComputer VisionGenerative ArtBrand IdentityCreative Tech

Spatial Brand Mirror

Your webcam feeds a generative art system that responds to your presence and motion in real time. The output is not a mirror — it is an abstract transformation. Movements become shape, colour, and rhythm on a p5.js canvas tuned to a specific visual palette. Diana builds this physical-digital feedback loop for experiential clients. This experiment shows it running in a browser tab, responding to whoever stands in front of it.

2026.06.17

EXP_010

p5.jsGenerative ArtOpenAIAgent OpsClaude DreamingParticle System

Agent Dream State Visualiser

Anthropic shipped Dreaming — between-session memory consolidation for Claude agents — in May 2026. This experiment shows what that process might look like as a generative canvas. A scrolling log feed from Diana's real agent fleet drives a background reflection cycle: every 12 seconds, a structured reflection arrives and the particle system shifts. Colour, velocity, and flow encode what the model noticed. The viewer reads that something is consolidating. They cannot name what.

2026.06.17

EXP_009

OpenAIAnthropicGoogleSSEReliability Benchmark

Agentic Reliability Dashboard

Head-to-head tool-calling reliability test across three frontier models. Run 15 tasks simultaneously against GPT-5.5, Claude Opus 4.7, and Gemini 3.1 Pro. Track pass rates, latency, and JSON validity in real time.

2026.04.26

EXP_008

LLMSSEGenerative UIA2UIStreamingAgent Ops

Generative UI Renderer

Describe a UI layout in plain text. An LLM generates a typed component tree and streams it live — each block renders as it arrives. Google's A2UI validates this pattern. Diana built it first.

2026.04.25

EXP_007

React FlowAgent ArchitectureOpenAISSEDrag-and-DropInteractive DiagrammingMulti-Agent Patterns

Orchestration Map

Drag-and-drop agent topology builder. You construct your own agent architecture, and an LLM annotates each node and edge as you build, interrogating your design decisions in real time.

2026.04.23

EXP_006

React FlowGoogle ADKAgent ArchitectureOpenAISSESequentialAgentMulti-Agent Patterns

ADK Visualizer

Static interactive diagram contrasting monolithic and orchestrated agent topologies from Google's ADK case study, annotated in real time by an LLM.

2026.04.21

EXP_005

Anthropic SDKClaudeBullMQSSERedisAgent OpsAutonomous Pipeline

Autonomous Brand Pipeline

Submit a creative brief and brand rules. A three-stage LLM pipeline generates N concept variants, evaluates each against your rules, and ranks the top K — streamed live as it runs.

2026.04.19

EXP_004

Anthropic RoutinesClaude CodeGitHub APISlackAgent Ops

Routines Repo Audit

Anthropic Routines powers a nightly Claude Code session that audits 8 repos for stale manifests, CI failures, open Dependabot PRs, uncommitted worktrees, and stale branches — then posts a structured summary to Slack.

2026.04.18

EXP_003

Canvas 2DBoidsSpatial HashGray-Scott RD

Crowd Flow Twin

This experiment couples three simulations: 10,000 autonomous agents running Boids flocking, a density trail grid tracking where agents have travelled, and a Gray-Scott reaction-diffusion system transforming raw crowd density into emergent organic patterns. Each subsystem reads and writes shared state. The result is a crowd simulation that visualises density as living structures, not heatmaps.

2026.04.15

EXP_002

Canvas 2DStable FluidsFluid DynamicsNavier-Stokes

Gesture Fluid Wall

This experiment runs a full Navier-Stokes fluid solver — diffusion, advection, pressure projection — at 60fps on a 256×256 grid using only CPU and Canvas 2D. No WebGPU, no WebGL. Stam's unconditionally stable semi-Lagrangian advection translates directly from offline VFX to interactive use. The simulation never blows up regardless of pointer speed.

2026.04.13

EXP_001