Skip to main content

Run a tree

Your agent runs a deployed tree one question at a time. SMEme sends a question and its options, your agent finds the answer in your files and apps, and SMEme decides what comes next. At the end you get a report with the outcome, the path that led there, and the evidence behind each answer.

Put the case where your agent can reach it

Attach the files, point to a folder, or name the connected apps that hold the facts, such as email, a CRM, or a shared drive. SMEme doesn’t store your files. Your agent reads them and sends each answer with a short reference to where it found it.

Run prompt

Run my [decision tree name] SMEme decision tree on this case: [one line describing the case].
The evidence is in [the attached files / this folder / these connected apps].

For each question SMEme asks, follow the hints on that question about where to look and what to avoid. Cite where each answer came from. If the evidence is missing or unclear, ask me. If I don't know either, leave it unanswered rather than guess. If two sources disagree, show me both before you answer.

When SMEme returns the report, tell me the outcome, the path that led there, and the evidence behind each answer.

What a run looks like

Reading the report

SMEme tells you when the answers aren’t enough, instead of guessing. Every result except Concluded tells you what to do next.

Result What it means What to do
Concluded One outcome follows from the answers. Act on it, or ask which answers decided it (see Ask more).
Several outcomes possible The answers so far don’t settle on one outcome. Answer the open questions, or ask what would settle it.
Needs more information Questions that decide the outcome are unanswered. Find the missing evidence and run again.
Answers don’t fit together Some answers can’t all be true under the tree’s rules. Recheck those answers and their sources.
Sources conflict The evidence disagrees about an answer. Decide which source to trust, then run again.

When the answers settle the outcome, the run returns a report. That report notes when the deciding answers were not independently re-checked.

Send a batch of answers in one shot

For a batch of cases, an audit, or an automated agent pipeline, your agent can send the answers it has for a case to a decision tree in one call, instead of one question at a time. SMEme reasons over the answers you send. Questions you leave out stay open, and the report tells you whether the outcome is settled without them.

Batch prompt

Load the worksheet for my [decision tree name] SMEme decision tree. For each case in [the attached files / this folder / this list], answer the questions you can from the evidence, following each question's hints on where to look and what to avoid, and cite where each answer came from. Leave a question unanswered when the evidence doesn't settle it.

Send each case's answers to SMEme in one call. Then give me a table with each case, its outcome, and any questions left open.

Next: Ask more →