---
title: Build Agents That Think, Act, and Learn
summary: Choose your LLM provider, set autonomy boundaries, define tool access, and deploy production-grade agents that handle real-world tasks end-to-end.
date: 2026-09-17
---

# Build Agents That Think, Act, and Learn

Choose your LLM provider, set autonomy boundaries, define tool access, and deploy production-grade agents that handle real-world tasks end-to-end.

## Multi-LLM support

Switch between Anthropic, OpenAI, Google, Amazon Bedrock, DeepSeek, Mistral, Groq, xAI, OpenRouter, Qwen, Moonshot AI, Ollama (local), and custom OpenAI-compatible endpoints. Each agent can use a different, independently-configured provider.

## Autonomy configuration

Three autonomy levels: **Supervised** approves every action before it runs, **Semi-auto** approves the plan up front and then executes it without further check-ins, and **Autonomous** runs end to end with no approval step at all.

## Templates

Six pre-built templates to start from: Customer Support Agent, Research Assistant, Content Creator, Lead Generation Agent, Personal Assistant, and Discovery Agent.

## Per-agent integrations

Each agent gets its own set of connected integrations, assigned individually, with no forced overlap between agents doing different jobs. See [Integrations](/en/docs/features/integrations) for the full catalogue.

## Scheduled execution

[Monitors](/en/docs/features/standing-orders) let an agent run autonomously on a schedule, not just in response to a message.

## Planning and delegation

A complex instruction becomes a [plan](/en/docs/features/planning) you approve before it runs, and any agent can [delegate](/en/docs/features/delegation) part of that plan to another agent it has access to.

## Technical specifications

- Maximum context window varies by model: current-generation models support up to roughly 1 million tokens; see [Models](/en/docs/guides/models) for what each provider offers
- 625 tools available via connected integrations
- Agent capacity depends on your plan: Trial allows 1 agent, Core allows 3, Scale is unlimited, and Nexus is set by your license
- Every configuration change is saved as a version automatically, for audit purposes

## Related

- [Core Concepts](/en/docs/guides/core-concepts): how agents relate to conversations, tools, and memory
- [First Setup](/en/docs/getting-started/first-setup): creating your first agent
- [Models](/en/docs/guides/models): choosing a provider and model for an agent