In Synapse you write each agent in plain language, saying what it does and how far it may go, and then you put several of them into a squad with a leader. The work arrives shared out, and the final answer comes back with the name of whoever wrote each part.
The specialist written in plain language, the folders and permissions it respects, the test with the hard case, and the squad that shows who did each part.
Drag the line and see the difference between explaining the process all over again and asking once, of people who already know it.
Six screens your team will use, in the order they show up: write the specialist, set its limits, put it in a squad and watch the work come back shared out.
You say what the agent does, how it answers and how far it may go. It is not code, it is the text you would write for someone on their first day, and you can try it with your hardest question before letting it out.
What the agent may look up, change or delete comes from a catalog organized by area, and every item says what it does. Whatever your role does not allow shows up blocked, so the agent never does for you what you could not do yourself.
A squad gathers the agents you wrote, with a scope they all share and one of them marked as leader. The leader goes in first and frames the task, and every agent after that already receives a summary of what came before.
In the squad panel you send the request and follow it member by member, who started, who finished and who failed, with the exact error message. Each test stands on its own, so the squad does not drag the previous conversation along.
The result comes back with each agent's contribution under its own name. When the answer came out of a document of yours it carries the file and the page, and when it is not in your sources the agent says so instead of making it up.
A squad can be the target of a support channel, become a step inside a workflow, or be fired at a time you set, by a webhook or by an event from the platform itself.
One agent per job, with the instructions and the reach you set.
Pick the members, mark who leads and write the scope that holds for all of them.
Watch each agent working and read the answer with the name of whoever wrote it.
Bring a process that passes through several hands today and we will show you, in your own context, how it looks shared out between agents.
Agents and squads belong to the workspace where they were created, and what an agent reaches is always what the person running it already reached.
On TVBluey, the company behind Synapse, on TECH SC by Globo NSC
An agent is one specialist with instructions of its own. A squad is a team of those agents, with a scope they all share and one of them marked as leader, who goes in first and frames the work.
No. A workspace starts with none, and you build yours out of the agents you wrote. That is why a squad ends up shaped like your process rather than like a generic template.
No. An agent's instructions are ordinary text, and what it may do you tick in a catalog organized by area.
No. The agent's reach is crossed with the reach of the person running it, so whatever your role does not allow shows up blocked in the configuration and never runs.
It does not have to. Each one has its own, chosen for the job, and the squad keeps a default model. The Multi-LLM layer explains how that choice is made.
The panel shows which member failed and the exact error message, and the others still come back with what they produced.
It can. A support channel can have the squad as its target, a workflow step can hand the task to it, and a webhook or a time you set can fire it.
Tell us which one it is and we will build the first squad with you, with the agents it would ask for.
Talk to us
