Introducing Kaer Workflows: Visual Pipelines for AI Agents

When we launched Kaer, agents worked one task at a time. You'd describe what you needed, pick a model, hit run, and get a result. For a lot of use cases, that's enough. But we kept hearing the same thing from teams using the platform: "I need agent A's output to feed into agent B, and I need it to happen automatically."

Today, we're launching Kaer Workflows — a visual pipeline builder that lets you chain agents, add decision logic, connect to external APIs, and run the whole thing on a schedule or in response to events.

Why workflows matter

Real work isn't a single task. It's a chain of decisions and actions that depend on each other. Consider something as routine as a weekly competitor analysis:

1. Search the web for recent news about three competitors
2. Summarize the key developments
3. Compare against your own product roadmap
4. Draft a brief for the leadership team
5. Send it via email every Monday morning

You could do each of these as a separate agent task. But then you're the glue — copying outputs, pasting them as inputs, triggering each step manually. Workflows eliminate the glue.

How it works

Workflows uses a canvas-based interface. You drag nodes onto the canvas, connect them, and configure each step. There are five types of nodes:

Trigger nodes start the workflow. They can fire on a cron schedule ("every Monday at 9am"), in response to a webhook ("when a new ticket is created"), or on-demand.

AI agent nodes are the workhorses. Each one runs an agent task — with its own prompt, model selection, and tool access. The agent receives the previous node's output as context.

Logic nodes add conditional branching. Route the workflow based on content, keywords, sentiment, or any condition you define. If the competitor analysis finds a major product launch, take path A (urgent alert). Otherwise, take path B (standard weekly brief).

Tool nodes connect to external systems. Send an email, post to Slack, make an HTTP request, write to a database. These are the connectors that bridge your agents to the rest of your stack.

Output nodes define where results go. Store them, email them, push them to a webhook endpoint, or display them in the Kaer dashboard.

What we got right

We tested Workflows with about 30 teams before this launch. A few things resonated immediately:

Each agent node is a full agent. This isn't a simplified "prompt + response" step. Each node gets full tool access — web search, code execution, file operations. An agent node in the middle of a workflow can search the web, write a Python script, execute it, and pass the results to the next node.

Error handling is built in. If a node fails — an API times out, a search returns nothing, the model produces a bad output — the workflow doesn't just crash. You can define retry logic, fallback paths, or skip-and-continue behavior per node.

It's visual, but not dumbed down. We know that "visual builder" sometimes means "limited." Workflows supports complex dependency graphs, parallel execution branches, and variable passing between nodes. If your use case is simple, the builder is simple. If it's complex, it can handle that too.

A real example

One of our early testers — a 15-person marketing agency — built a workflow that processes new client briefs. Here's what it does:

1. Trigger: Webhook fires when a new brief lands in their project management tool
2. Agent 1: Reads the brief, extracts key requirements, and identifies the target audience
3. Agent 2: Researches the competitive landscape for that audience
4. Agent 3: Drafts an initial content strategy based on the brief + research
5. Logic node: If the budget is above $10K, routes to a senior review queue; otherwise, routes directly to the team lead
6. Output: Posts the strategy doc to their shared workspace and notifies the assigned team

The whole workflow runs in about three minutes. Previously, this process took their team about two hours of manual work per brief.

What's next

Workflows is available today on all Pro and Enterprise plans. Starter plan users can build workflows with up to three nodes — enough to get a feel for what's possible.

We're already working on workflow templates, version history, and collaborative editing. If you build something interesting, we'd love to see it — reach out to us on the community page.