Agentish Framework
A Practical Guide to Building Workflows
This guide teaches you how to use the Agentish visual editor to build, configure, and export agent workflows. Each chapter covers one aspect of the framework — from individual nodes to complete multi-agent systems.
Chapters
Overview
What is Agentish, what is a workflow, the visual editor UI, and the design → export → compile → run pipeline.
The Entry Point
Every workflow starts here. Why the entry node matters, how to define initial state variables, and what happens at startup.
The LLM Node
The core building block. System and human prompts, input/output state, structured output, tool assignment, and the terminal node rule.
The Router Node
Conditional branching powered by an LLM. Router values, output slots, and the minimum two-output requirement.
The Worker Node
Subtask delegation. How workers attach to LLM nodes, their fixed return format, and when to use them.
Connecting Nodes
Edges, slots, and wiring. NormalEdge vs ConditionalEdge, building pipelines, fan-outs, and hierarchies.
Loops
How to create proper iterative workflows. Loop mode (fresh vs continue), feedback state variables, exit conditions, and common pitfalls.
State
The most important concept. Global state schema, data types, how LLMs communicate through state, and how to read execution results.
Tools
The Function Catalog, MCP tools, custom tools, tool assignment, and iteration limits.
Logging & Observability
Understanding what your agent did. Langfuse traces, reading execution logs, and debugging failed runs.
Troubleshooting
Common problems and how to fix them. Infinite loops, recursion limits, prompt quality, and compilation errors.