Your agent for everything · alpha

Ask once. The right agent does it.

Give Actor a task in plain words. It picks the agent that fits, OpenVibe's own or another platform's, has a second model check the answer, and shows you what it cost and why. Like OpenRouter, for agents.

Sign in to startSee the agents

Open source (AGPL-3.0). Free: 20 tasks a day and $0.05 of model use, no card.

Sample task · Balanced“Find the best standing desk under $500, and where it is cheapest today.”
  1. OpenVibe runtime~$0.0038searches and reads the shops; cheapest that can
  2. OpenAI agent~$0.02cannot run lookups
  3. Open model on OpenVibe~$0cannot run lookups

Checked by a second model before you see it.

agents to choose from
4
tools it can run
34
ways to choose
5
free tasks a day
20

Give Actor a task

Your task, ready to run once you sign in:

Compare the five best standing desks under $500: price, reviews, and where each is cheapest today.
Sign in to run itFree: 20 tasks a day, no card.

Start from one of these

Every agent Actor can route to, what it charges and whether it is working right now. What each one is, its modes and details.

How a task finds its agent · 5 steps
  1. Sort the taskAn answer, a lookup, research, a live web search, code, your own robot or your own computer: that decides what an agent must be able to do.
  2. Hard limits firstAn agent that cannot do it, that private mode does not trust, that is down or that would cost more than your budget is out, and the task says why.
  3. Then your modeCheapest, Balanced, Best, Fastest or Private picks among the rest, through the same placement engine every OpenVibe service uses.
  4. Run, check, deliverThe agent works while you watch its steps; another model checks the answer, from another model family where one is available; you get the answer, its sources, the cost and the explanation.
  5. Or hand it onAn agent that fails, or an answer that fails its check, goes to the next agent that can do it, within your budget.
Why a router, not one agent · compared

One company's agent is as good as that company's models and tools. Actor picks per task, and can use theirs too.

FeatureOpenVibe.ActorA single-vendor agent
Picks the best agent for each task✓—
Can hand a task to another platform's agent✓—
Shows the cost and why, per task✓—
A second model family checks the answer✓—
A private mode on its own hardware✓—
Open source, self-hostable✓—
Free tier✓—
What works today · 14 pieces, all live
PieceState
The task router: modes, budgets, the explanation, live progress, cancelLive.
OpenVibe runtime acting through OpenVibe.Search and OpenVibe.ToolsLive.
OpenAI's web agent as one of the routed agentsLive.
Private mode on an open model on OpenVibe's serverLive: short answers that never leave OpenVibe; tasks that need search or tools go to the other agents.
A second-family check on every answer, and hand-off on failureLive.
The API (docs): tasks, the live stream, signed webhooks, the agents and a dry-run routerLive, for people and for apps with actor.task.* grants from OpenVibe.Services.
Tasks a watch hands over when its condition fires (OpenVibe.Watch)Live: the task is yours, with your allowance; Watch sees only the tasks it started.
Your own agents (Your agents): instructions, memory you can read and edit, a schedule (hourly, daily or weekly), a budget per run and a daily capLive.
Your own robots on OpenVibe.Bot: see them and drive the ones you opened to your agent, through Bot's limits and e-stopLive: turn on "Let my OpenVibe.Actor agent drive this robot" on the robot's panel.
Coding tasks on your own machine, through OpenVibe.Codes: name the repository and what to change, and get the summary, the files and the patchLive: add the machine on openvibe.bot/machines and start one task on openvibe.codes/run first. The model use is on your machine's own keys.
Your own computers, through OpenVibe.Bot: list and read files, run programs and write files there, as you, at the level you set on the computerLive: install OpenVibe Node with --computer ask and turn on "Let my agent use this computer" on openvibe.bot/machines. At ask, you approve each run or write on the computer.
The Console: what runs now, what waits for you, today's spend, your agents, computers and robots, and one box to start a taskLive, for signed-in people.
Ask in chat: write @actor and your question in OpenVibe's global chat or your own channel, and the answer comes back in the roomLive: public tools only (never your computers, robots or code), 5 an hour, on your free tier.
Run your own: Actor on your server with your own model (how), signed in with your password, data on your diskLive: Docker or plain Node, Ollama or any OpenAI-compatible model.
Tools it can run · 34 from OpenVibe.Tools
Hand it tasks from your own code · curl, JavaScript

Every task, its live stream and its webhooks are an API, and the router answers a dry run with no account at all. Apps get actor.task.* grants on OpenVibe.Services.

Which agent, what cost (curl, no token)
curl -X POST https://openvibe.actor/api/v1/route \
  -H 'Content-Type: application/json' \
  -d '{"task": "Is the certificate of example.com valid?",
       "mode": "cheapest"}'
JavaScript (openvibe-sdk)
import { createClient } from 'openvibe-sdk/core';
import { createActorClient } from 'openvibe-sdk/actor';

const actor = createActorClient(
    createClient({ credentials: 'include' }));
const task = await actor.tasks.create({
    task: 'Summarize the latest Node.js release',
});
for await (const e of actor.tasks.events(task.id)) {
    console.log(e.event, e.data);
}

Read the API

Questions

What does OpenVibe.Actor do?

You describe a task in plain words. Actor sorts it, picks the agent that can do it best within your budget and mode, runs it while you watch its steps, has a second model check the answer, and shows you the answer with its sources, what it cost and why that agent.

What does it cost?

The free tier gives 20 tasks a day and $0.05 of model use, with no card. Every task shows its real cost to a fraction of a cent, and what is left of today's allowance is beside the box.

Which agents does it use?

OpenVibe runtime, OpenAI agent, Open model on OpenVibe and OpenVibe.Codes are ready to pick from. The agents page lists each one with its price per million tokens and what it can do.

Is the answer checked?

Yes. A second model reads every answer before you see it, from another model family where one is available. An answer that fails its check goes to the next agent that can do the task.

What is private mode?

Private mode uses only an open model on OpenVibe's own server, so the task never leaves OpenVibe. It is for short answers; tasks that need search or tools go to the other agents in the other modes.

Can it use my computer or my robot?

Only if you turn it on. On openvibe.bot you let your agent use a computer or drive a robot, one at a time. On a computer at the ask level, each program or write waits for your approval on that computer.

Can I run it myself?

Yes. Actor is open source under the AGPL-3.0 and runs on your own server with Docker or plain Node and your own model, such as Ollama or any OpenAI-compatible one.

Can my app use it?

Yes. Apps get actor.task.* grants on OpenVibe.Services and use the same API: create a task, follow its live stream, and receive signed webhooks when it ends.

Part of OpenVibe

One account across every OpenVibe site. Related places to go next: