Manage agents (Flows)

Manage agents (Flows)

Flows is Anter's agent configuration and execution environment. Define an agent's model, instructions, tools, and skills — then run it from the chat interface, the API, or on a schedule. For complex tasks, compose multiple specialized agents that hand work off to one another.

Core concepts

ConceptDescription
AgentThe configured AI instance — model, system prompt, tools, skills, and permission settings
ProjectOrganizational scope that groups agents, issues API keys, and manages tool connections
Sub-agentA specialized agent that an orchestrator can transfer control to or delegate tasks to
SkillA domain instruction bundle that loads on-demand to extend an agent's capabilities for a specific topic
RunA live execution — streams back responses and preserves a full routing trace

How it works

1

Create a project

A project is the unit of organization — it groups your agents together and is the scope for API keys and tool connections. Start by creating one for your use case.

2

Build an agent

Define the agent's model, system prompt, and permission mode. The permission mode controls whether the agent acts autonomously or pauses for approval before taking sensitive actions.

3

Attach tools and skills

Connect tools — HTTP webhooks or MCP servers — to let the agent take action in external systems. Add skills to give the agent deep, domain-specific instructions that load on-demand as the conversation requires them.

4

Add sub-agents (optional)

For workflows that span multiple domains, define sub-agents under an orchestrating agent. The orchestrator can transfer control to a sub-agent entirely, or delegate a task and collect the result before continuing.

5

Run the agent

Start a run from the chat interface, call the streaming API, or set a cron schedule for unattended execution. Every run streams events in real time and records a full routing timeline you can inspect afterward.

When to use Flows

Flows works well when you need:

  • Conversational AI without building your own orchestration — the agent loop, tool dispatch, and memory management are handled for you
  • Multi-agent coordination — route queries to specialized agents using transfer (one winner) or delegation (fan-out and collect)
  • External integrations — call webhooks or connect MCP servers to give agents access to your systems
  • Multiple LLM providers — run agents on OpenAI, Anthropic, Azure OpenAI, Azure AI Foundry, Google Gemini, or NVIDIA models, from a single interface
  • Scheduled or API-triggered runs — beyond interactive chat, agents can run on a cron schedule or be invoked programmatically

New to Anter?

Start with the Quickstart — you'll have a working agent running in under ten minutes.