Agent Builder
Build a custom AI agent — a configurable assistant with its own persona, memory, knowledge base, tools, and deployment surface — separate from generating a full app.
What You Can Configure
| Capability | Description |
|---|---|
| Persona & model | System prompt, temperature, and choice of underlying model |
| Persistent memory | The agent extracts and remembers facts across conversations, scoped globally or per user |
| Knowledge base (RAG) | Upload files (PDF/DOCX/XLSX/CSV/TXT); the agent searches them to answer questions |
| Tools | Web search, calculator, file search, delegate-to-another-agent, connector queries, payment links, and more |
| Connectors | OAuth-connect Gmail, Sheets, Slack, GitHub, Notion — the agent can query your connected data |
| Scheduled tasks | Cron-based background tasks the agent runs on its own |
Deploying an Agent
Once built, an agent can be:
- Chatted with directly in the dashboard
- Embedded on any website via a self-contained widget script (floating bubble + iframe)
- Listed on the marketplace for discovery
- Called via API or the MCP integration (
create_agent,chat_with_agent)
Multi-Agent Chaining
An agent can delegate part of a conversation to another agent (delegate_to_agent), useful
for splitting a complex task across specialists. Delegation depth is capped to prevent
infinite recursion.
Agent Builder vs. AI Team
These are two different things that are easy to conflate:
| Agent Builder | AI Team (staffing) | |
|---|---|---|
| What it is | A standalone custom assistant you design from scratch | A pre-configured team hired onto a specific app |
| Use when | You want a chatbot/assistant for a use case that isn't a full app | You've already built an app and want agents operating it |
| See also | This page | AI Team & Loops |
Getting Started
Create an agent from the dashboard's Agent Builder, or via the MCP integration's
create_agent tool from ChatGPT, Claude, or Gemini.