Agents & smart routing

Agents you calllike any other model.

An agent bundles a model with a system prompt, tools, knowledge bases and MCP servers. It gets its own model name — so every OpenAI-compatible client can use it.

Start for free Knowledge bases

No base fee · No minimum term · Pause any time

What an agent combines

Model and promptHosted or commercial, with your system prompt
Knowledge basesAnswers grounded in your documents
Tools via MCPConnect external tools and APIs
Routing or workflowRules or a visual editor

What you get

From simple assistant to routed workflow.

Start with one prompt and grow it into rules or a visual workflow — without changing how your application calls it.

Own model name

Each agent is addressed by its own name through the OpenAI-compatible endpoint.

Knowledge and tools

Attach knowledge bases and MCP servers so the agent can look things up and act.

Rule-based routing

Route by token count, keywords, regular expressions, tool use or attached knowledge to the model that fits.

Visual workflows

Build multi-step flows in a visual editor instead of writing orchestration code.

Cost-aware

Keep routine requests on cheaper or hosted models and escalate only when needed.

Optional response cache

Reuse answers to similar questions and skip the model call altogether.

How it works

Build an agent in three steps.

Create the agent

Choose a model and write the system prompt.

Attach knowledge and tools

Add knowledge bases and MCP servers from your workspace.

Call it by name

Use the agent name as the model in any OpenAI-compatible client.

Business model

Costs follow usage.

An agent adds no fee of its own. You pay for what it uses.

Pricing

The models it calls

Hosted models by runtime, commercial models per token.

Hosted models it relies on

Billed by the minute while they run — stop them when nobody needs them.

Attached knowledge bases

$0.05 per hour plus $0.01 per GB and hour.

Keep exploring

Common questions

Questions about agents.

A configured model with a system prompt and optional tools, knowledge bases, MCP servers and routing. You call it by name like any other model.

You define rules — for example by token count, keywords or tool use — or let the automatic router choose. Decisions are recorded per request.

No. Agents are configured in the dashboard, including the visual workflow editor.

Yes. Rules can send requests to either, which is how you keep routine work cheap.

Ready to get started?

Build your first agent.

Create a free account, pick a model and give it knowledge and tools.

Start for free Pricing