Skip to content

Standard, Deep analysis, and Build

Every exploration runs in one of three modes. The mode shapes Ronja’s tools and how much rigor she applies, so choose it at the start.

On a fresh, empty exploration, the owner sees a mode control with up to three segments: Standard, Build (shown only if you have build access), and Deep analysis.

The exploration start screen with the Standard, Build, and Deep analysis mode segments and helper copy visible The mode picker appears only on an empty exploration, and only for its owner.

Once the first message is sent, the mode is fixed: a later change is refused with “mode can only be changed before the first message”. The one exception is that Ronja can switch a Standard conversation to Build (see Ways in); nothing becomes Deep analysis after the first message. Deep analysis and Build conversations carry a header stamp, “DEEP ANALYSIS” or “BUILDER”.

The default, and the mode for most explorations. Ronja answers questions, queries and charts your data, and explains what she finds. If you have build access, she can also start a one-click sign-in to connect a service such as Google Sheets or Fortnox without leaving Standard; setting up a database or other credentials moves the conversation to Build. A substantive answer arrives as a titled, laid-out answer (see The laid-out answer); a short exchange comes back as plain text, and so does a turn that closes on an inline app, when the useful answer is something to work rather than read.

For questions where you need to trust the answer: “A rigorous, methodology-first investigation — Ronja designs the approach, validates the data can support a trustworthy answer, and explains how it reaches every finding.” It uses the same tools as Standard, apart from connecting a service, and doesn’t build resources, but reasons at the deepest level.

  1. Ronja writes the Methodology, a living document covering the goal, hypotheses, method, feasibility and data quality, and validation. Its panel opens on her first write and stays reachable from the exploration’s rail.
  2. She normally hands the conversation back to you before running the analysis. Read the Methodology, and reply to go ahead or change the approach.
  3. The conclusion arrives as a laid-out answer, with the estimate and how sure Ronja is of it at the top, the evidence beside the sentence that reads it, and the assumptions and limitations under it. Where the finding is something to work, such as a sensitivity you move a slider on, it can arrive as an inline app instead, with the finding, the method and its limits above the frame: one or the other, never both, and never in place of the Methodology check-in.

Deep analysis is available to everyone, but only chosen up front: on the start screen, or as the mode of a tangent.

In Build mode Ronja designs and creates durable resources such as tables, workflows, apps and notes. She writes an implementation Plan first and keeps it up to date, and a collapsible build panel above the message box tracks what is in progress and what has been built. The header stamps “BUILDER” with the subtitle “extended thinking”. Build replies are plain text: the structure lives in the build panel and its Plan.

Everyone has build access unless an Admin with billing access turns it off for their role or for them on the Control Center’s Rules tab. Super Admins always have it, and User Read-Only members never do. Without it, the Build segment doesn’t appear, Ronja can’t start a sign-in to connect a service for you, and the server refuses with “Build mode is disabled for your account. Contact an administrator to request access.” Deep analysis doesn’t need build access. See Control Center.

  1. The start screen: pick Build before the first message.
  2. Mid-conversation: Ronja switches a Standard conversation to Build herself when it turns into a build job, including when a recurring ask would be better as an automation. She says so and carries straight on, with no approval step; build access is still required. The conversation stays in Build mode from then on.
  3. A tangent: when starting a tangent, you pick its mode; Build is offered if you have access.
  4. A button on a resource opens a Build conversation focused on it: Edit with Ronja and Update with Ronja on an app, Investigate with Ronja on a failing table build, workflow or automation, and Edit in chat on a Saved Agent.

The first thing Ronja does in Build mode is plan. She writes a short implementation plan in the build panel, which opens by itself, and uses it to confirm the scope with you before creating anything.

Read the plan and steer it: correct a wrong assumption, cut what you didn’t ask for, add a constraint. It is the highest-leverage moment of a Build session, because Ronja then builds against a scope you agreed on, and Build mode is the most credit-intensive work in Ronja, so a wrong turn is cheapest to fix in the plan.

The Build-mode build panel showing Ronja’s short implementation plan above the message box Ronja writes the plan first — read it and steer it before she starts building.

When the plan is right, say go in one of three ways:

You Ronja
Click Approve & build in the build panel Starts building, and for the rest of the conversation creates, edits and deletes resources in the feature the plan builds in without asking each time.
Click Build, but ask me each time Starts building, records no approval, and keeps checking with you before each consequential step.
Tell her to go in your own words Starts building. The buttons are a shortcut for that message, not a gate in front of it.

If your reply changes the plan instead, such as a document to follow, new numbers or a correction, Ronja revises the plan, tells you what changed, and waits for you to say go.

Open the plan to read which feature that is, and what the approval does and doesn’t cover, before you click. After a click both buttons stand down and the panel says which kind of build is running. While the approval is live, Withdraw approval ends it, and Ronja’s next step asks again.

What Approve & build covers:

  • Deleting inside that feature, because replacing a table she got wrong is part of building. What she deletes goes to the Trash for 30 days; see Restore from trash.
  • Only the feature the plan names. If the plan also points at something you can’t open, or at someone else’s private work, even as an admin who can open it, she says so and keeps asking before touching it.
  • Not sending email, deleting a whole feature or clearing a table, which always ask, and not review: on a shared resource your changes still travel as drafts and proposals through an administrator. See Versions, drafts, and approvals.

If the plan doesn’t name a feature, either button starts the build, but Ronja checks with you before each change either way, and the plan says so. Tell her where it should live, and approve again for the standing approval. If Ronja rewrites the plan while you read it, approving is refused with a note that the plan changed, nothing is built, and the panel reloads.

Once you’ve said go, the panel becomes a progress board. A plan with more than a step or two carries a build-steps checklist that ticks off as each step lands. Collapsed, the card shows one line: the current step, how far along she is, and whether her checks are green. Its footer holds the session History (“History · N changes this session”) and a link to the full plan.

For more habits, see Getting the most out of Ronja.

Build mode changes what Ronja can make, not who approves it. Consequential actions still pause on the approval strip, and edits to shared resources travel as drafts and proposals through review. See Versions, drafts, and approvals.