Use preset AI agents in Chat
An agent saves you from rewriting the same long setup prompt every week. This page shows you how to open a chat from a catalog agent, how to create your own, and exactly what an agent is and is not on this surface.
What is an agent here?
On chat.synapsex.ai an agent is a saved persona: a system prompt plus a set of capabilities that starts an ordinary conversation. That is the whole mechanism.
Nothing runs on a server on your behalf. Starting an agent opens a normal chat with that agent’s prompt applied. There is no background execution, no scheduled run, and nothing is billed per run.
That honesty matters because it tells you what agents are good for: repeatable framing. If you always want log analysis done the same way, an agent gets you there in one click instead of one paragraph.
Browse the catalog
Open Agents in the sidebar, or go to chat.synapsex.ai/agents . You can:
- Filter by category — document, devops, research and others
- Search with the search box, which matches agent names, descriptions and tags
Each card shows what the agent is for and which capabilities it declares.
Use in Chat opens a new conversation with that agent’s persona applied. From there it behaves like any other chat: your model choice, your attachments, your history.
Create your own agent
Click Create Agent on the Agents page. That opens a separate form at
/agents/new, where you provide:
| Field | What it is for |
|---|---|
| Agent Name | How it appears in your catalog |
| Description | What it is for |
| Category | Where it files in the catalog |
| Default Model | Which model a chat with this agent opens on |
| System Instructions / Prompt | The actual instructions — this is the agent |
| Capabilities & Tools | Optional toggles for Web Search, Code Interpreter, File Search and Quantum Access |
The Default Model list here is its own short list, not the six-model catalog you get in the composer’s model picker. Pick from what the dropdown offers, and change the model inside the conversation afterwards if you want one that is not on it.
Custom agents are stored on your account, so they follow you to another device.
Worked example
Create an agent named Release note writer with this system prompt:
You turn raw commit logs into release notes for a technical audience.
Rules:
- Group changes under: Added, Changed, Fixed, Removed. Omit empty groups.
- One line per change, imperative mood, no commit hashes.
- Collapse mechanical commits ("fix typo", "bump deps") into a single
"Maintenance" line.
- Never invent a change that is not in the input. If the log is ambiguous,
list it under "Needs review".Now paste a commit log into a chat started from that agent, and paste the same log into a plain new chat with no agent.
The plain chat gives you a summary in whatever shape the model felt like. The agent chat gives you four fixed headings, imperative one-liners, and anything ambiguous parked under Needs review instead of guessed at. Same model, same input — the difference is the saved prompt.
Which model should an agent use?
Match the model to the work, exactly as you would for a normal chat: Forge for code-heavy agents, Oracle for reasoning-heavy analysis, Atlas when the agent runs often and the work is routine. The comparison table is in Choosing a model.
The Default Model you set on an agent is only the model the conversation opens on. Once you are in the chat you can switch to any model in the picker with Cmd+Shift+P, so an agent never locks you out of a model.
Related
- Custom skills — for rules that should apply to every chat, not one agent.
- Memory and personalization — the profile that applies underneath any agent.