Weft lets you design, rehearse, and run AI-powered agent workflows. You compose agents on a visual canvas, connect the tools your team already uses, and let the workflows coordinate multi-step tasks that used to need manual glue code.

Create a workspace, drag your first agent onto the canvas, and pick a trigger. The starter templates cover the common patterns — a summarizer, a reviewer, and a support triage flow — so most teams ship a working workflow in their first session.

Weft ships native connectors for chat, code hosting, ticketing, and data warehouses, plus a generic webhook node. Anything with an HTTP API can join a workflow, and the Connector SDK lets you package internal services as first-class nodes.

Yes. Workflow state is encrypted at rest and in transit, agents run with least-privilege scopes you grant explicitly, and simulation runs never leave your workspace. Audit logs record every trigger, tool call, and outcome.

Every workflow has a sandbox mode: replay historical events, step through agent decisions, and inspect intermediate outputs before promoting to production. Simulations are free to re-run while you iterate.

Automated steps execute as soon as their inputs are ready; manual steps pause the workflow and wait for a human approval. You can mix both in one canvas — most teams keep a manual gate before any externally visible action.