Skip to content
Ronja Docs

Tangents

A tangent lets you branch off an exploration to chase a side question — without cluttering or risking the main thread. The original conversation is untouched; the tangent is a private copy that’s entirely yours.

Forking

Section titled “Forking”

Hover any assistant reply and click the branch icon — “Explore a tangent from here”. A menu titled “Explore a tangent in…” offers the branch’s mode: Standard, Build (if you have build access), or Deep analysis. Pick one, and you land in a new exploration containing a full copy of the conversation up to that reply.

Three things to know about the copy:

  • Inherited history is visually de-emphasized, ending in a divider: “Branched from {parent} — history above is inherited”. Everything below it is the tangent’s own.
  • Tangents are born untitled. The branch names itself from its own first prompt, just like any exploration — and until you’ve prompted it, it appears only in the branch navigation, not your sidebar lists.
  • A Deep-analysis tangent starts with a blank Methodology — it designs its own investigation rather than inheriting the parent’s.

You only need view access to the source to fork — so you can branch privately off a teammate’s shared exploration without touching their thread.

The branch breadcrumb pills at the top of a tangent — one pill per branch, with the current branch filled solid and the branch-tree icon beside them Inside a tangent: the breadcrumb of branch pills sits above the thread, and the Summarize & return call-to-action waits at the bottom.

Section titled “Navigating a family of branches”

Once an exploration has more than one branch, a breadcrumb of lane-colored pills floats above the thread — one per branch you can open, with the current one filled solid. A pill pulses while its branch is streaming and bolds when it has unseen output.

For the full picture, click the branch-tree icon next to the pills to open the Branches panel: a git-style graph of the whole family, one dot per turn, with dashed curves marking returns. Clicking a dot jumps to that turn. In the parent, forked-from messages carry a marker — “Tangent branched from here” — with chips that open each tangent.

The Branches panel open beside a chat, showing a git-style graph of the tangent family with the current branch marked “here” The Branches panel: a git-style graph of every branch in the family, so you can see where each one split off and jump between them.

Branch views are access-filtered: branches you can’t open (someone else’s private tangent, for example) are simply not shown.

Folding findings back in

Section titled “Folding findings back in”

When a tangent pays off, bring the result home. Above the composer — shown when you can write to the parent — sits “Done exploring this tangent?” with the button “Summarize & return”.

The modal drafts a summary for you: Ronja re-reads the tangent and writes a handoff report (“Reading back through this tangent to draft a summary…”). You then edit it — toggle Edit/Preview — and, if the family has several ancestors, choose the “Fold into” target. Click “Fold in”.

The summary lands in the target as a collapsed card headed “Merged: {tangent title}”, with a status badge, a “View branch” link back to the tangent, and the full report behind a chevron. The main conversation’s agent can build directly on it.

Nothing forces a return, though. A tangent that didn’t pan out can simply be left — or deleted like any exploration.