Asks the product, not you
The setup meeting a new tool usually needs is one call your agent makes on its own. It comes back knowing what it may do here and what it may not, so the first thing you are asked for is a decision, not a configuration.
How it works
The agent operates it. You inspect it.
The usual arrangement is a CRM a person maintains with an AI feature bolted on to help. Indraft runs the other way round: the agent is the operator, and the application is your lens on what it did.
What the agent does
Everything a person would have had to do first
There is no setup phase to get through. Your agent shapes the workspace and fills it in before you have opened it, and the first thing you see is a CRM with your customers already in it.
Shapes the workspace
Sets up a pipeline with stages that match how the business actually sells, and defines typed custom fields where they earn their place.
Fills it in
Companies, contacts, and opportunities, interactions with their source, and follow-up tasks it can read back later, from the context it already has.
Resolves its own ambiguity
When an upsert could match two records it gets the candidates back and chooses, or creates a distinct record with an explicit override. It is never allowed to guess.
What you do
Read it, correct it, undo it
Open the activity view and see, in order, what changed and who changed it. Click a value for its previous value, the source it came from, and whether it was observed, inferred, imported, or supplied by a person.
Fix what is wrong. Undo what should not have happened. That is the whole job, and on most days it takes minutes.
Anything it gets wrong, you can take back
Undo a single field from the activity view, or unmerge two records that should never have been joined. Where something genuinely cannot be taken back, it says so before you click rather than after.
The boundary
What Indraft leaves to you and your agent
- It makes no model calls. There is no assistant inside Indraft competing with yours, and no inference line on your CRM bill. Your agent supplies the judgment; Indraft validates it, stores it, and shows who asserted it.
- It takes no raw custody. It does not connect to your inbox or your calendar, and it stores no message bodies, transcripts, or attachments. Your agent reads those and writes the conclusion, with its name on it.
- It has no workflow engine. No automation builder, no forecasting, no dashboards. The thing that decides what happens next is the agent you already run.
Next
See what your agent builds
Connect the agent you already run and give it an afternoon. What you get back is a working CRM, not a sandbox with sample data in it.
Starting from an existing CRM
A workspace starts empty. There is no importer: you bring the records in, either by pointing an agent at the system that holds them or by writing them through the bulk endpoint, a hundred at a time. Whichever way they arrive, every row carries where it came from, and you can take the whole workspace back out from Settings whenever you want.