How does each::sense pick a model?

each::sense reads your prompt’s intent and matches it against the catalog, weighing each model’s capabilities, cost, latency, and quality. It routes to the single best fit, or, when the task needs several models, assembles and runs them as one multi-step workflow.

Which modalities does each::sense support?

each::sense covers image, video, and audio generation, plus image editing, across the models in the each::labs catalog. It can take attached media as input and factors it in when choosing the right model for your task.

Which languages does each::sense understand?

each::sense understands natural-language prompts across many languages through frontier LLMs. English, Turkish, Spanish, German, Japanese, and Korean prompts all work for both model selection and execution.

all platform products

LIVE NOWOPENAI-COMPATIBLE

Describe it. sense picks the models and builds the workflow.

each::sense is the natural language agent in front of every model in the catalog. Tell it what you want, an image, a video, a voiceover, a 3D asset, and it chooses the models, chains them into a workflow, and runs it. OpenAI compatible base URL; drop into LangChain, CrewAI, AutoGen, or any agent SDK you already ship.

try the agent

Tell us your AI chaos. We’ll show you the way out.

Type your actual problem. Our AI agent reads it, gives you a short answer, and links you straight to the part of the docs that solves it. No signup, no sales call.

600+

models · auto-picked

model names to remember

OpenAI

compatible API

modalities · image · video · audio · 3D

⚐ what the agent does the moment you call it

An agent that owns model selection, so you don’t have to.

sense reads your request in plain English, decides which models can land it, chains them into the right order, and runs the pipeline. The model names, the failure surfaces, the cost trade-offs, all hidden behind one OpenAI compatible endpoint.

mechanism01

›prompt → models · auto

Natural language requests

"Make a 6-second product video with a voiceover" is the API. sense parses intent, maps to the right combo of image, video, and TTS models, and runs them, without you naming any of them.

mechanism02

›one prompt · N steps · 1 trace

Automatic workflow construction

Multi step pipelines get built on the fly. Image → upscale → animate → narrate is one request to sense; one workflow under the hood, one trace, automatic recovery on failure.

mechanism03

›base_url: eachsense-agent.../v1

OpenAI compatible drop in

Point any OpenAI SDK at https://eachsense-agent.core.eachlabs.run/v1 and start calling. Works with LangChain, CrewAI, AutoGen, and every agent framework you already wired up. Zero new SDK to learn.

Change the base URL. Keep the SDK.

sense ships an OpenAI-shaped endpoint, so anything that already speaks OpenAI, official SDK, LangChain, CrewAI, AutoGen, the Vercel AI SDK, becomes a media-aware agent the moment you point it at the new URL.

example · openai sdk

import OpenAI from "openai";

// each::sense is OpenAI compatible.
// Point any OpenAI client (or LangChain / CrewAI /
// AutoGen / Vercel AI SDK) at this base URL.
const client = new OpenAI({
  apiKey:  process.env.EACHLABS_API_KEY,
  baseURL: "https://eachsense-agent.core.eachlabs.run/v1",
});

const response = await client.chat.completions.create({
  model: "each::sense", // sense decides which media models to run
  messages: [
    { role: "user", content: "make me a 6s product video with a voiceover" },
  ],
});

// behind the scenes:
//   sense → wan-2.5 (video)
//         → suno-v3 (voice)
//         → compose → 1 trace · $0.19

WORKS WITH

OpenAI SDKLangChainCrewAIAutoGenVercel AI SDKAnthropic SDK

BASE URL

https://eachsense-agent.core.eachlabs.run/v1

Use it with model: "each::sense". Tool calls and streaming behave exactly as your framework expects.

Reach for sense when…

usertierregion

You don’t want to be the one picking models

Six hundred models, each with its own quirks. sense knows which ones land on your prompt and picks them by quality × latency × cost. You ship; the model panel is sense’s problem now.

600 → auto · top-1

You need a pipeline, not a single call

Story → storyboard → video → narration → cut. That is four model calls and a join. sense builds the workflow from one prompt and runs it as one trace, with each::workflows handling versioning and rollback underneath.

one prompt → workflow

kling-v3wan-2.7

You’re already on an agent framework

Wired into LangChain, CrewAI, AutoGen, Vercel AI SDK, or anything OpenAI-shaped? Change one base URL; sense becomes the model behind every tool call. Streaming and tool-use work exactly as your framework expects.

base_url → swap

You want to ship without picking a winner

New product, no benchmarks yet. sense’s choice is good-enough on day 1 and gets better as the catalog ships new models, your code stays the same.

catalog grows · code static

Other products you’ll use alongside this.

PLATFORM

sense builds the workflow; workflows give you version, rollback, and the trace.

PLATFORM

Every model sense picks runs through the router; spills happen for free.

PLATFORM

eachenhancer

Prompts that sense ships get enhancer-polished on the way in, lower error rates.

* FAQ

FAQ

01 / 04

What is each::sense?

each::sense is a natural-language AI agent that decides for you. Describe what you want in plain language, whether an image, a video, or a voiceover, and sense picks the right models from the each::labs catalog, chains them into a workflow, and runs it. You don’t choose a model or a modality; sense does.

LIVE NOWOPENAI-COMPATIBLE

Stop picking models. Ship the request.

OpenAI compatible drop in. Usage-based pricing, you only pay for the models sense runs.

Follow the white rabbit See sense docs