June 2026

When you allowlist an operation on a tool-provider node, you can now pin it to specific resource IDs. Expand Limit to specific... under an operation and pick which documents, sheets, or other records the agent may act on. When a create operation makes a new resource, its ID is added to the matching scope for you, so the agent can immediately read back what it just created.

The Add tool palette on an agent node now lists the Alarm, Filesystem, and Tool nodes alongside integrations and MCP server. Pick one and it is created, wired into the agent, and opened to its own config step, so you no longer have to drop the node on the canvas and draw the edge by hand.

An agent that runs in a sandbox now always has an upload_file tool that publishes a file from the sandbox to a public URL, so it can share a generated image, attach a report to an email, or pass a file to a downstream node. The five CLI harnesses always run in a sandbox, and the built-in agent runs in one whenever a filesystem node or a repository mount is wired, so you no longer need a filesystem node just to publish a file.

The Feed has a new Agents tab that lists every tool call your agents make, grouped by run, with the agent's model, each call's inputs and result, and the final response. Click a credential to open it in the editor.

Connect Discord by signing in and installing the NoClick bot into your server, with no keys to copy. You can then start a workflow when an app event happens: your bot is added to or removed from a server, a premium entitlement is created, updated, or deleted, or a slash command is run. Wire the trigger into an AI agent, or add the Discord node as an agent tool provider so the agent can run Discord actions for you.

Wire the inbound email trigger to an agent and the agent can reply in the same thread, securely scoped to that conversation. A new send email node delivers notifications to your inbox, formatted from markdown, with a one-click disable link.

Wire any integration node into an AI agent and the agent can call its operations directly, like creating a Linear issue or sending a Slack message. You choose exactly which operations the agent is allowed to use, and every tool call shows up live in chat and in run replays.

Wire a trigger (Telegram, Slack, email, or an alarm) into an agent and the incoming event becomes the agent's message, with conversations threaded per channel. Pair it with send operations to build a reply bot entirely on the canvas.