Custom Agents

Build your own agent.
Wire it to your own systems.

Agent Architect is a visual, no-code designer for creating your own AI agents — pick the model, write the system prompt, attach tools that call your own APIs and databases, and deploy it as a reusable node in the Workflow Editor.

How You Build One Integrate Your Systems Deploy & Reuse
How You Build One

Visual configuration,
no code required

Agent Architect lets you design a custom agent from scratch or from a template. You define every aspect — which model backs it, what its system prompt says, and what it's allowed to read and call — and it's ready to use immediately.

  • Model — pick any active cloud model (Claude, GPT-4o) or local model (TinyLlama, Mistral)
  • System prompt — define the agent's role, output format, and constraints in plain language
  • Knowledge base — link collections so the agent retrieves and cites the right context automatically
  • Test in Prompt Lab first — iterate on the prompt before saving an agent version
Read Agent Architect Docs
totalapp.app / agent-architect
New Agent · Draft Claude Sonnet
"You are a returns-eligibility checker. Always respond in JSON with keys: eligible, reason."
Tools attached: check_order_status, get_return_policy
i Knowledge base linked: Returns & Warranty Policy
Save Agent → appears immediately as a node in the Workflow Editor.
No restart or re-deployment required
Integrate Your Systems

Give the agent access
to your own tools

The Tool / Function Registry is how a custom agent reaches outside TotalApp — an HTTP call to your own API, a SQL query against a connected database, or a call to another TotalApp workflow. Define a tool once, attach it to any agent.

  • Name, description, and a JSON Schema parameter spec — the model reads these to decide when to call it
  • Implementation options: your own HTTP endpoint, a SQL query, a sandboxed JS function, or another workflow
  • No manual prompt engineering — TotalApp injects the tool schema into the agent automatically
  • Every call is logged in Observability, with the exact arguments and the value returned
Read Tool Registry Docs
totalapp.app / tool-registry
Tool Registry — New Tool
check_order_status(order_id: string) → your internal API
Implementation: HTTP request to api.yourcompany.com/orders
Attached to: Returns Eligibility Agent
Call logged in Observability — args and return value both visible for debugging.
Reusable across any agent you build
Deploy & Reuse

Save once, drop it
into any workflow

Click Save Agent and your custom agent becomes an immediately draggable node in the Workflow Editor's palette — no restart, no re-deployment. It behaves exactly like a pre-built agent type, because it's built on the same architecture.

  • Appears in the Workflow Editor node palette under "Custom Agents" the moment it's saved
  • Every save creates a new version — running workflows keep using their pinned version
  • Upgrade a workflow to a newer agent version explicitly, whenever you choose
  • Reuse the same agent across as many workflows and screens as you need
Open Workflow Editor
totalapp.app / workflow-editor
Node Palette — Custom Agents
Returns Eligibility Agent — v3
Dragged onto canvas — ready to wire to Order Status trigger
Two other workflows still run on v2 until you explicitly upgrade them.
Live in the palette — no redeploy needed
Everything You Configure

The full agent-building toolkit

Every piece your custom agent needs, all inside TotalApp — no separate infrastructure to stand up.

Any Model, Cloud or Local

Back your agent with Claude, GPT-4o, or a local model like TinyLlama or Mistral — swap models without rebuilding the agent.

Prompt Lab for Iteration

Test and refine the system prompt in a live sandbox before it becomes a saved agent version.

Tool / Function Registry

Give your agent access to your own APIs, databases, or other TotalApp workflows via defined, reusable tools.

Knowledge Base Linking

Attach document collections so the agent retrieves and cites relevant context automatically for every query.

Workflow Editor Node

Every saved agent shows up as a draggable node — wire it into triggers, branches, and other agents like any other block.

Full Observability

Every run, tool call, and token is logged — audit exactly what your agent did and why.