Feature · Multi-Agent Workflows

Compose agents that talk to each other

Visual editor. 18+ node types including triggers, agents, conditional logic, parallel execution, integrations, and Computer Use. Build the workflows your agency needs in hours, not weeks.

What it does

Most AI tools give you a single agent and hope it can do everything. KILN lets you compose multiple agents into a workflow — a triage agent qualifies, a research agent enriches, an action agent executes. Each step is its own node, with its own model, prompt, and config.

The editor is visual. You drag a node onto the canvas, connect it to the next, and inspect the data shape between nodes. Conditional branches handle “if hot lead, route to CRM; else, send a nurture email”. Parallel branches execute simultaneously and merge — for fan-out / fan-in patterns like “summarise from three sources, then synthesize”.

Every workflow run is observable. Inputs, outputs, and decisions at each node are logged with a clickable timeline. Failed steps can be retried or replaced without restarting the whole flow.

For agencies, this is the difference between “custom build per client” and “template that scales”.

How It Works

Three moving parts

Step 01

Drop nodes onto the canvas

Triggers (webhook, schedule, lead, chat). Agents (chat, task). Logic (if/else, switch, filter). Actions (HTTP, email, Slack). Integrations (Sheets, Notion, Airtable). Computer Use, RAG, sub-workflows.

Step 02

Wire data between nodes

Each node exposes a JSON output schema. Map fields explicitly or pass payloads through. Schema mismatch warnings highlight broken connections before run-time.

Step 03

Run, observe, iterate

Hit Run. Watch the live execution timeline. Inspect inputs and outputs at every step. Bugs surface as failed nodes you can replay individually.

Use Cases

What agencies actually build with this

Lead Qualification Pipeline

Trigger on new lead → score with an LLM → branch hot vs cold → route to CRM, email, or Slack accordingly.

Customer Support Tiers

Greeter agent triages → diagnose with the knowledge base → either resolve directly or escalate to a human handoff.

Content Generation

Research agent gathers sources → drafter writes → editor agent revises → publish to CMS via integration node.

Technical details

  • 18+ node types: triggers, agents, logic, actions, integrations, AI tools, advanced (sub-workflow, agent-team, ensemble)
  • Visual editor with drag-and-drop, multi-select, copy-paste, undo/redo
  • Conditional branches, parallel fan-out, merge-back, sub-workflows for reuse
  • Per-node retry policies, dead-letter queue, deterministic replay
  • Live cost tracking, token counts, and per-node duration metrics
  • Export workflows as JSON for version-control and migration

Ready to build with this?

Free forever for testing. Start charging your first client by next week.

Built in GermanyGDPREU-hosted