An assistant that actually reads your workflow

Most "AI features" in automation tools are a text box that writes generic advice. The chat panel in Pipory's editor does something narrower and more useful: it reads the workflow you have open and answers questions about that workflow.

What it can do today

Open any workflow and ask the panel things like:

• "What does this workflow do?"
• "Why is this node configured this way?"
• "What ends up in the context after the loop finishes?"

The assistant receives the actual graph: the nodes, the edges between them, and each node's configuration. It answers from what is really there, not from a guess about what workflows usually look like. Credentials are redacted before the graph is sent to the model, so your keys and connection strings never leave your box.

It runs on an Anthropic credential you provide, the same bring-your-own-key model as the LLM nodes. Your assistant, your key, your spend.

Why "understand" came first

We sequenced the assistant deliberately. Explaining an existing workflow is the safest place to start: the assistant only reads, so a wrong answer costs you a shrug instead of a broken automation. It is also the foundation the next steps need, because an assistant that can edit a workflow must first reliably understand one.

What comes next

Two more capabilities are on the roadmap, in order:

• Build and edit. Describe a change in plain language - "when a Stripe payment comes in, post to Slack" - and get a proposed edit rendered as a diff on the canvas. Nothing applies until you click apply.
• Debug. The per-node execution logs already record every input, output, error, and duration. The debug step lets the assistant read those logs for a failed run and explain which node failed and why, offering the fix as a proposed edit when the cause is a configuration problem.

The through-line: AI in Pipory is not a separate product surface. It is the same canvas, the same graph, and the same logs you already work with, made conversational.

If you are new to the canvas, build your first workflow and then ask the assistant to explain it back to you. It is a good test of both.