Skip to content

All Tutorials

Step-by-step paths through the documentation, from a first agent to multi-agent systems. Each step links to the page with the full details and code.


1. Your First Agent

Build, run and deploy an agent that answers tasks.

  1. Prepare an environment: Install & Setup.
  2. Write the agent class with on_preprocess and on_data: Writing the Agent Code.
  3. Unit-test it, then run it locally and send it a task over WebSocket: Testing Agents.
  4. Register its specification: Agent Registration.
  5. Deploy it to a cluster and check it with /healthz and /dryRun: Agent Deployment and Smoke-Testing.

2. Give the Agent a Brain and Tools

  1. Connect a model: Agent × AI Model Interface.
  2. Add tools and functions: Agentic Function & Tool Calling.
  3. Add memory: Agentic Memory.
  4. Stream tokens and tool calls to a UI while it works: Live Task Streams.

3. A Multi-Agent Pipeline

  1. Learn the workflow types: Introduction to Workflows.
  2. Build, register and run the five-agent Legal Contract Review Pipeline.
  3. Make it adaptive with a router agent: Dynamic Workflows.

4. Agents That Check and Improve Their Work

  1. Describe a good result as an expectation and evaluate outputs: Evaluation.
  2. Let the agent iterate until it passes: Self-Evaluation Loops.
  3. Add a human approval step for risky answers: Human-in-the-Loop.
  4. Turn outcomes into lessons: Loop Improvement.

5. Agents That Organise Other Agents

  1. Choose between synchronous and asynchronous collaboration: Agentic Patterns.
  2. Spawn dedicated sub-agents for a task, then remove them: Spawning Sub-Agents.
  3. Let an LLM plan a workflow from a goal and replan on failure: Prebuilt Operators.
  4. Recruit specialists from a market: Task Exchange & Bidding.

6. Skills and Generated Code

  1. Write a subjective skill and load it on demand: Subjective Skills.
  2. Write an objective skill and run it with the interpreter: Objective Skills.
  3. Generate, test and reuse code as a tool: Running Generated Code in the Ecosystem.

7. Bring an Existing Agent

Run a LangChain, LangGraph or other framework agent on the platform: Plural Agentic Backends.

8. Production

  1. Attach governance and guardrail policies: Agentic Governance.
  2. Auto-scale and survive failures: Scaling.
  3. Checkpoint long-running tasks: Persistence, Checkpointing & Recovery.
  4. Watch it run: Observability.

Complete Examples

Example Shows
Legal Contract Review Pipeline A multi-agent static workflow, end to end
Advanced Memory End-to-End Example Memories, the Learner and the Incorporator working together