​
Agents

Agents are custom AI personas your team can build and share. Each agent has a name, a system prompt, a chosen model, and optionally a set of tools it can use. Team members can start conversations directly with an agent, or invoke one in a team channel with an @mention. Agents are scoped to a team — Studio Editors and Admins manage them from Settings → Agents, and team members can use any agent shared with their team.

​
Creating an agent

1

Open Agents

Navigate to Settings → Agents and click New Agent.
2

Configure the basics

Give the agent a name, a description, and write a system prompt that defines its behavior and persona. Select the model it should use by default.
3

Add tools (optional)

Enable any tools the agent should have access to — HTTP requests to external APIs, MCP server tools, or knowledge base search. See the sections below for details.
4

Save and share

Save the agent. It immediately becomes available to your team in the agent selector when starting a new chat.

​
Using an agent in chat

1

Select an agent

When starting a new chat, click the Agent dropdown (or the agent icon in the sidebar) and choose the agent you want to talk to.
2

Chat normally

The agent’s system prompt is applied automatically. You interact with it the same way as any chat — the agent handles tool calls and knowledge retrieval behind the scenes.

​
HTTP tool use

Agents can call external APIs during conversations. When the model decides to use the http_request tool, ManyLayers executes the call from the server and feeds the result back to the model automatically — no manual steps required. Your admin controls which external hosts agents are allowed to reach: 
suite:
  http_tool_allowlist:
    - api.example.com       # exact host match
    - .internal.corp        # matches all subdomains of .internal.corp
An empty allowlist blocks all outbound HTTP tool requests. Contact your admin if an agent needs access to a host that isn’t listed.

​
MCP tools

Agents can use tools from registered MCP (Model Context Protocol) servers. Your admin registers shared MCP servers that are available to all agents. You can also register personal MCP servers from your profile settings — these are only visible to agents you create. Browse available MCP servers in Settings → MCP Servers. See MCP Gateway for details on how MCP integration works.

​
Prompts

Prompts are reusable system prompt templates with full version history. Use them to store common instructions and share them across multiple agents, or use them as conversation starters. Manage your prompts from Settings → Prompts. Every time you update a prompt, the previous version is saved — click Version History on any prompt to view or restore an older version.

​
Memories

Memories are per-user snippets that agents can recall across conversations. When you save something as a memory, it persists and the agent can reference it in future sessions — useful for things like your preferred coding style, project context, or personal preferences. Manage your memories from Profile → Memories. You can add, edit, or delete any memory at any time. If a semantic embedding model is configured, memories are also searchable by meaning — not just exact text — so the agent can surface the most relevant context automatically.

​
Channel @agent mentions

Invoke any agent in a team channel by typing @agentname in a message. ManyLayers matches the mention to the agent, runs the inference in the background, and posts the reply as a channel message visible to all channel members. The original message is saved immediately; the agent’s reply appears asynchronously when it’s ready. See Collaboration for more on channels.