Get started
How Sands works
Sands is an AI agent. Every request becomes a run: Sands decides the steps, reads and acts in your tools, checks the results, and shows each step as it happens.
The run
When you send a message, Sands works out what you are asking for, picks the tools it needs, and does the work. The reply opens with one line for that work, like Worked for 12s. Expand it to see each step: what Sands read, what it found, and what it changed. The answer streams in below.
Each step says what it did and closes with what it found. When Sands checks your board, the step shows which project it read and the numbers it got. When it creates a ticket, the step shows the ticket and links to it.
Why show the work
You should never have to take a claim on trust. If a number or a change matters, the evidence for it is one click away in the run.
Reads and writes
Reads
Reading is how Sands grounds its answers: listing projects, fetching issues, pulling worklogs, reading a Slack thread. Reads never change anything, and Sands does them whenever an answer needs live facts.
Writes
A write changes something in a connected tool: a new ticket, an assignment, a comment, a Slack message, a plan sent as a project. Every write follows the same rules:
- It must be something you asked for. Sands never invents a change on its own.
- It acts on the exact item you meant. If Sands is not sure which ticket or channel you mean, it asks.
- Your approval mode decides whether it runs right away or waits for your OK.
- After it runs, Sands reads the result back and reports what actually happened.
- It runs once. Retrying a request or re-sending a plan never duplicates the change.
What Sands writes in your name
Text that Sands posts to your tools, like a ticket comment, is either your own words or facts it read from your tools, such as when a ticket was last updated. When a change carries text, the approval card shows the full text before you approve it.
Artifacts
Some replies produce something you keep working on: a plan, a report, an uploaded file. These are artifacts. Plans open in a drawer beside the chat and stay editable; every artifact in a conversation is listed under the Artifacts button at the top of the chat. See Projects and outputs.
Stopping a run
While Sands is working, the send button becomes a stop button. Stopping halts the run on the server, not just the animation. A run you stop before Sands replies or uses a tool costs no credits.
When Sands acts
Sands works when you ask, in a chat or in Slack. It does not run on a schedule, watch your boards in the background, or start work on its own.
What Sands remembers
Within a conversation, Sands keeps the thread: what you asked, the plan you are working on, and what it read. It also remembers which project tool and project you work in, so you do not need to repeat them. Starting a new chat starts a new thread.