Loading...
Loading...
Into the beautiful unknown.
Building sandcastles with
Notes, datasets, and findings from our ongoing studies, backed by downloadable data. Each piece is open, structured, and ready to build on.
Evals·12 Jul 2026
We measured 13 multi-agent patterns against one-model controls, with token costs and injected faults. Most cost more and score less; verification decides containment.
Workflows·03 Jul 2026
A catalogue of 75 generative media pipelines across image, video, audio-driven animation, and 3D, with sourced stages and explicit controls.
Agents·21 Jun 2026
Compare 41 hosted runtimes, agent SDKs, code sandboxes, and browser services, drawn from a source-backed catalogue of 95 AI infrastructure products.
Agents·01 Jun 2026
Compare 61 coding-agent capabilities across 28 tools, including plan mode, subagents, hooks, memory, MCP, and checkpoints. Browse all 1,708 coded cells.
Agents·04 Apr 2026
A source-backed catalogue of 40 ways LLM agents coordinate, from role-based workflows and verification loops to debate, shared state, and adaptive topology.
Evals·12 Jul 2026
We measured 13 multi-agent patterns against one-model controls, with token costs and injected faults. Most cost more and score less; verification decides containment.
Workflows·03 Jul 2026
A catalogue of 75 generative media pipelines across image, video, audio-driven animation, and 3D, with sourced stages and explicit controls.
Agents·21 Jun 2026
Compare 41 hosted runtimes, agent SDKs, code sandboxes, and browser services, drawn from a source-backed catalogue of 95 AI infrastructure products.
Agents·01 Jun 2026
Compare 61 coding-agent capabilities across 28 tools, including plan mode, subagents, hooks, memory, MCP, and checkpoints. Browse all 1,708 coded cells.
Agents·04 Apr 2026
A source-backed catalogue of 40 ways LLM agents coordinate, from role-based workflows and verification loops to debate, shared state, and adaptive topology.
The future does not arrive finished. You find the pattern. You build what comes next.
Three ways to build. Wire multimodal pipelines node by node, seat a table of agents on a shared canvas, or spin up a Data Lagoon that fills on repeat and exports like Airtable.
39+ nodes spanning image, video, audio, and language. Connect them visually, swap models per step, and run the whole chain from one canvas.
Multiple AI models take turns on a shared canvas, building on each other's work with no central coordinator. Six collaboration modes, from spatial reasoning to narrative development.
How it works
Turn-based collaboration
Agents observe the canvas, reason about shared state, and take actions in turn.
Any model, any slot
Mix providers freely. Claude for reasoning, GPT for creativity, Gemini for planning.
Plugin architecture
Each game defines its own rules, agents, and canvas. The engine handles the rest.
Live streaming
Watch every agent action arrive in real time. No waiting for a final result.
6 Collaboration Modes
Train Tracks
Build connected track on a tile grid with 4 agents
Writers Room
Phased narrative development with creative conflict
Brick Builder
LDraw 3D builds with critic gates and spatial queries
Podcast Studio
Multi-speaker podcast production with TTS synthesis
Model Arena
Head-to-head model competition with ELO ratings
Pattern Lab
13 interaction patterns, measurable on real tasks
Lists go stale. A Data Lagoon compounds: define schema and cadence once, then agents discover, validate, and merge rows on schedule. Seed from our catalog or start empty. Run it via MCP, query over REST, export when ready.
{
"mcpServers": {
"hermosa": {
"type": "streamable-http",
"url": "https://hermosaai.com/mcp",
"headers": {
"Authorization": "Bearer hermosa_key_…"
}
}
}
}table name
LA Specialty Roasters
entity label
Roaster
slug
la-specialty-roasters
seed from · optional
or start empty · upload CSV · point agents at URLs
schema
cadence
Set once. Waves run on schedule until you change it.
CSV
export snapshots
SQL
query live data
API
build on top
The models everyone uses, plus the cult favorites
One key for workflows, collab, and your Data Lagoon. Run agents over REST, query verified knowledge, or wire Lagoon MCP into the editor you already use.
Workflows & collab
Run visual pipelines and multi-agent sessions programmatically over REST.
Knowledge
Query verified catalog topics and your Data Lagoon snapshots. Export rows any time.
Lagoon MCP
Point Cursor or Claude Code at the Lagoon server to run discover, validate, and merge waves.
Prepaid credits
Buy a balance, spend it on any model or agent. One key, no per-call billing surprises.
x402 micropayments
Pay per request over HTTP 402. No account needed, no minimum spend.