Search that improves itself, with autonomy you control.
SWIRL 5.1 puts your team's judgment to work. Curation decisions become tuning proposals with the evidence attached, routine version elections can run on their own inside guardrails you set, and the assistant lets you choose how much research each question deserves. Here is an early look at three features, each with a short demo.
SWIRL already finds every version of a document; auto-pinning decides which one wins, at the level of autonomy you choose: Off, Assist, or Auto. In Auto mode, SWIRL elects and pins the best version of a document during the search itself - but only when every guardrail passes, including a clear winning margin and cluster integrity. Ambiguous cases wait for a human in the review queue.
Every curator decision becomes training data. The nightly improvement pipeline analyzes rejection telemetry and proposes precise, bounded tuning changes, with the supporting evidence attached to each proposal. One click applies a change live with no restart; one click reverts it.
Not every question deserves the same effort, so you choose per question, right in the assistant composer. Quick runs one search and returns one cited answer. Thoughtful runs multiple rounds of live federated search from different angles, then delivers a single synthesized answer that tells you how many rounds it took.
Deploy SWIRL Enterprise privately in your own cloud workspace, or email us and we'll stand it up against a slice of your systems in a 30-minute working session - on your own data.