Goals
Set a goal in the project chat, and agents work through the project's studies, documents, conversations and notes, then post an answer that cites its sources.
Preview, Wrengler Pro. Goals run on the platform, so you can close the chat while one works. The answer lands in the conversation you set it in.
A goal is something you want found out about a project, in your own words. For example: find the typologies in this project from the sunlight studies, the design documents and what we've said about them. Agents plan the work, read the project's own data, and write the answer up with its sources, like an agent in a chat app. A goal only reads. It never changes your project's data or your Rhino model.
Set a goal
In the project chat, in Rhino or on the web, press + and choose Goal, or choose Set a goal… from the command palette. Write what you're after and send it. The goal becomes your message in the conversation, with its card under it.
Good goals name what to find and where to look:
- Find the typologies in this project, with how much sun each one gets.
- Which of our studies contradict the design statement?
- Summarise every decision we made about the courtyard, with where each was made.
What a goal reads
Only the project it was set in, and only what you can see:
| Kind of data | What the agents read | Counts as |
|---|---|---|
| Facts | the project card, its tags, workspace-approved knowledge | recorded facts |
| Studies | saved sunlight, view and radiation studies: settings and summary results per group | recorded facts |
| Documents | the project's files, passage by passage, with file and page | what a document says |
| Conversations | the project's chats: what people asked and what the assistant answered | what people said |
| Notes | the project's notes, and your own | what people said |
Archived files and deleted conversations are left out. Text from files and people is read as information, never as instructions: a document can't tell a goal what to do.
How it works
- Plan. A planner looks at what the project holds and splits the goal into up to four parts, each reading one or two kinds of data. If it can't make a plan, the goal runs as one part that may read everything.
- Work. A worker per part, two at a time, reads what its part needs: a list first, then the items that matter. Every item it reads gets a reference such as S3.
- Write. A writer turns the findings into the answer: first the answer itself, then the detail (often a table), then what the project's data didn't have. Every claim cites its source.
Numbers
A number in the answer comes only from a recorded fact: a study's result, a tag, or an agreed workspace fact. It's quoted as it's stored. A number that only a document or a person gives is attributed to them: the design statement gives 24 m.
After the answer is written, Wrengler checks every number in it against the facts the goal read. Any number it can't find there is listed under the answer, so you know to check it:
Not found in this project's recorded facts: 24. Check these against the cited documents before relying on them.
The answer
The answer is posted in the conversation like any other, with its Sources listed under it: studies, files (with the page) and conversations, each opening where it came from. You can rate it, and it becomes part of the conversation. Ask a follow-up in the same thread and the assistant builds on it.
If the write-up itself fails, the answer is each part's findings as they came, so the work isn't lost.
Stop and resume
- Stop ends the goal after the step in hand. The person who set it and the project's leads can stop it.
- If the platform stops before the goal finishes, the card says Interrupted. The person who set it can press Resume. The plan and every finished part are kept, and only the part that was in progress starts over.
Limits
| Limit | Value |
|---|---|
| Parts per goal | up to 4, two at a time |
| Reads per goal | 48 |
| Time per goal | 4 minutes of work and writing |
| Model spend per goal | about $1 |
| Goals you can set | 10 an hour |
| Goals running at once | 2 per person |
| Length of a goal | 2,000 characters |
When a goal runs short of time or spend, it writes the answer from what it has and names the parts it couldn't finish.
Goals come with Wrengler Pro. They count against your organization's monthly assistant messages, and an organization that has turned AI features off can't run them.
Not yet
- Goals across projects. A goal reads the project it was set in.
- Changing things. A goal can't add tags, save files or change the model. It tells you what it found, and you decide.
- Comments on files and results. What people say is read from the project's conversations and notes.
When something goes wrong
| Message | What to do |
|---|---|
| Two of your goals are already running. | Wait for one to finish, or stop it. |
| You can set 10 goals an hour. | Try again later. |
| Goals are a Wrengler Pro feature. | Goals need Pro. |
| Your organization has turned AI features off. | Ask an organization admin. |
| Interrupted — the platform stopped before this goal finished. | Press Resume, or set the goal again. |
Related
- Assistant: chips, sources and privacy.
- Harness Overview: how the assistant acts in Rhino.
- Projects: the files, facts and conversations a goal reads.