Agentivity
Star on GitHub Start free
EN FR
How it works The whole picture: describe what you need to the Assistant, or build it by hand in the Studio. No code. See how it works →

Build

TeamsAI specialists that work together AgentsOne specialist, one role, its own tools WorkflowsFixed steps, with a person who approves

Make it better

Prompt StudioTest and improve what your agents are told Knowledge Bases & Data TablesYour documents and tables, for your agents

Use it

VibeAppsApps your users simply talk to AssistantBuild by describing what you need
Ready-made teams on Synergi →Install the free Community edition →
Start free
EN FR

Product · Agents

One AI specialist. Brief it like a colleague.

An agent is one AI specialist with a role. You describe what it is good at and how it should work, in plain words. Then you give it the tools and the knowledge the job needs.

Start free Build it with the Assistant
Recorded in the Studio, played faster.

How you build one

Three things. The rest only when the job needs it.

An agent with a role, good instructions and a model is already useful.

  1. A role: one or two sentences on what it is good at
  2. Instructions: what it must do, and how, in plain words
  3. A model: the AI model it thinks with, from your own provider
The agent Reminder Writer in the Studio: the agent box, with its AI model, a chat channel, a session and the Payment terms knowledge base plugged in, and its role in the panel on the right
Reminder Writer: it writes a polite payment reminder, and checks the payment terms before it does.

Give it what the job needs

Tools, knowledge, a memory, a person to ask.

Each one plugs into the agent on the canvas. Add only what the job asks for.

  • Tools A calculator, a web search, a data table. The agent chooses which one to use.
  • Your knowledge Your own texts, in a Knowledge Base. The agent receives only the passages that fit its question. Knowledge & Data
  • A memory A session keeps the conversation, so the agent remembers what was said before.
  • A person to ask A channel to reach a person when the agent needs an answer.
  • Guardrails Checks on what goes in and out of the agent, for example to block a word.
  • A way of working Answer directly, use its tools, or think, act, look at the result and repeat until it can answer.

In the Studio

What you see when you build one.

The agent Invoice Clerk in the Studio, with its AI model, a chat channel, a session and three tools: a data table, a calculator and a document reader
An agent and its tools Invoice Clerk reads the supplier invoices you upload, files each one and checks its figures.
The window of an AI Agent step: the way it works, the request it receives and its instructions, written in plain words
Its instructions, in plain words What it must do, and how. Like a short brief for a new colleague.

Where an agent works

Alone, in a team, or in a workflow.

  • Alone You open it and give it a task. It answers, and uses its tools when it needs them.
  • In a team It becomes a member, next to other specialists. The team reads its role to decide when to use it. Teams
  • In a workflow It becomes one step among fixed steps. AI works only where you put it. Workflows

Questions

What people ask us.

Do I need to code?

No. You write the role and the instructions in plain words, and you plug in what the agent needs on the canvas. Or you describe the agent to the Assistant.

Which AI models can an agent use?

The one you connect, with your own key: Anthropic, OpenAI, Google Gemini, xAI, DeepSeek or AWS Bedrock.

When should I use a team instead?

When the job has several parts that need different skills, or a check by someone else. One agent is enough for one clear task.

Is it free?

The Community edition is free forever, on your own server.

Your next hire is an AI team.

Start free in the cloud: nothing to install.

Start free

14 days free · no credit card

Hosting it yourself? The Community edition is free. Install Community