Skip to content

Policies

A policy is a short document of standing rules for how Ronja should work. Ronja is given it at the start of every conversation, so a rule you write once applies to every answer after it — without you repeating it.

There are two documents, and every conversation carries both. The organization policy can also hold a short policy for each value of a people property, which only the people with that value get:

Policy Who writes it Who it applies to Cap
Your policy You Your own conversations 10,000 characters
Organization policy Admins Every conversation and every automation in the organization 20,000 characters
Policy for a value, such as Policy for CEO Admins Conversations with the people who have that value 500 characters each, kept with the organization policy

Where they disagree, the organization policy’s organization-wide rules win over everything else, and your own policy wins over the policies for your properties. See Policies for people properties.

Open the account menu and click Your policy. The page holds one document, yours alone: nobody else’s answers change because of it, and nobody else reads it.

The kinds of rule that belong here are the ones you would otherwise repeat in every chat:

  • Report amounts in SEK.
  • Prefer tables over charts unless I ask for a chart.
  • For revenue questions, read the revenue definitions note first.

Admins reach it in the admin sidebar under Data & Knowledge → Organization policy — not the Organization rules page, which holds settings. Edit the document under Standing rules; expand What belongs in a policy? for guidance. It holds the rules that are true for the whole company:

  • The fiscal year runs January to December.
  • Exclude test accounts from every customer count.
  • For definitions of our revenue metrics, read the revenue definitions note first.

Only admins can edit it. It is applied to everyone’s conversations rather than shown to them in the app, so most people never read the rules their answers follow. Keep it short and true, and prefer a rule that points at a document over a rule that tries to be the document.

Each value of a people property can have its own short policy telling Ronja how to work with the people who have that value. Give Job role: CEO the policy “Lead with the answer and the number; no code unless asked”, and every CEO’s conversations follow it; a developer’s don’t. Like everything in a policy, it shapes how Ronja works, never what anyone can see or do.

A value’s policy applies wherever a person with that value is asking: their chats in the app, and their messages to Ronja in Slack or Microsoft Teams once their account there is linked to Ronja. It never applies to an automation or a Saved Agent run, which get the organization-wide rules only.

When rules disagree, Ronja follows them in this order:

  1. The organization policy’s organization-wide rules.
  2. The person’s own policy.
  3. The policies for their property values. They rank equally with each other: when two disagree (say Job role: CEO and Region: North), Ronja follows the one that fits the question and says so when that is unclear.

So a value’s policy is a default for people like them, and anyone can change it for themselves by writing the opposite in Your policy.

An admin writes a value’s policy on the value itself, on its property’s page in People properties. Each value shows its policy, or the line “No policy. Tell Ronja how to work with people who have Technician.” with Add policy. Clicking Add policy or Edit policy opens Policy for Technician:

  1. Write a few lines. Headings are not allowed; Markdown and references are. Click Insert reference to point at a note, table, feature or another resource.
  2. Watch the counter, such as 113 / 500. Over 500 characters it adds how many to remove, and Save policy is disabled.
  3. Check who it reaches: “Applies to 5 people: Erik Lund, Omar Haddad, Sara Ek and 2 more.”
  4. Click Save policy.

If the text references something some of those people can’t open, the dialog warns you, without blocking the save: share the resource with them, or accept that Ronja can’t read it in their chats.

To remove a value’s policy, empty the text and click Save policy. Every change to a value’s policy shows up in the organization policy’s History. Deleting a value keeps its policy, which then applies to nobody until you remove it or add the value again.

The Organization policy page lists every value’s policy under Policies for people properties, with the value, the policy’s text and how many people have the value. To add or change one, open the property in People properties.

A policy whose value or property was deleted reads “Applies to nobody: …”. Click Remove it, then Remove policy, to delete it, or add the value (and its property) again by the same name to make it apply again.

Under Preview for a person, choose someone under Person to see what Ronja is told about how to work with them: the organization-wide rules, then each policy for their values with its reason, such as “because their Job role is CEO”, and which of their values have no policy. Their own policy also applies, but is never shown to you.

Your policy shows each person the value policies that apply to them, under Also applied to you, each with its reason, such as Job role: CEO · because your Job role is CEO. Nobody sees there the policies for values they don’t have.

The organization policy reaches work done from outside Ronja as well:

  • ronja context, the command a coding agent runs after signing in with the Ronja CLI, prints the organization-wide rules first, then only the value policies that apply to the signed-in person, then that person’s own policy. A developer’s coding agent is never handed the CEOs’ policy, and nobody else’s agent is handed yours.
  • An AI assistant connected to Ronja reads the same rules, your own policy included.

So a rule such as “every table name starts with the source system” is read by Claude Code, Codex or any other agent building against your organization, not only by Ronja.

Every member can read the whole document through the CLI or the API; only an admin can change it there.

An admin changes it with their own personal access token (the kind ronja login creates), and ronja context shows an admin the exact commands, which change one part at a time so they never remove a value’s policy by accident. A token from the admin API Tokens page can’t write it, and neither can Ronja from a Saved Agent or an automation. A coding agent working under your token is told to change it only when you ask, and to show you the new text first; History then records the change as Edited by hand, by you.

A policy holds rules for how to work. It is not the place to record what your data means or what happened in a conversation. Each line should be one of two things:

  • Unconditional — “Report amounts in SEK.”
  • A pointer — “For revenue questions, read the revenue definitions note first.”

What does not belong: what happened in a conversation, a number, a status, or anything that will be out of date next week. Every future conversation reads the whole document as the present, so a line about last quarter’s migration will be read as if the migration were happening now.

Knowledge itself lives elsewhere — in notes inside a feature, in the data dictionary, in a codex. A policy line names the place and says when to read it. See How Ronja learns for the whole picture.

  1. Click Edit. On the organization policy, the editor holds the organization-wide rules only; each value’s policy is edited on its property’s page.
  2. Write your rules. Markdown is supported, and the counter under the box shows how much of the cap you have used — you cannot save over it.
  3. To point at something in Ronja, click Insert reference, search for the note, table, feature or another resource by name, and pick it. The reference goes in as a link Ronja can open, and it keeps working when the resource is renamed.
  4. Click Save (or press Cmd/Ctrl+Enter; Esc cancels).

Each save makes a new version, listed under History with when it was made and whether Ronja wrote it or it was edited by hand. On the organization policy, each entry also names the part that changed, such as Edited by hand · Policy for Job role: CEO. History shows the 20 most recent versions, and the last 50 are kept. Click Restore on one to bring it back as the newest version; the oldest entry, Empty — before anything was written, takes a document back to nothing.

On the organization policy, restoring brings back the whole document, value policies included. When that would remove a value’s policy or bring one back, Restore says so first and restores only when you click Restore again.

If the document changed while you had it open, because Ronja wrote to it or another admin saved the organization-wide rules, your save is refused and you see what is stored now:

  • Keep my text writes yours over it.
  • Load the stored version brings theirs into the editor and keeps your text below, so you can merge by hand.

If the organization policy can’t refresh while you edit, your draft stays open; Try again reloads the saved version without saving your draft.

Tell Ronja a lasting preference in a conversation, such as “always report in SEK”, and she can add it to your policy herself, for anyone, User Read-Only members included. A card in the chat reads Policy updated, with the lines she added or removed and an Undo button; click it and the card reads Undone. If the policy has moved on since, Undo says so and links you to the document rather than wiping what is there.

A change to the organization policy stops and asks first:

  • Ronja shows an approval request reading “Change the organization policy — every run in the organization is told this”, with the changes highlighted and the character count against the cap. Nothing is written until an admin approves it, and the approval has to come from a person signed in to Ronja, not a chat driven from the CLI or a coding agent (see Actions the API can’t take).
  • A User Read-Only member who asks gets no approval request: the change is refused.
  • Ronja changes only its organization-wide rules. If what she writes carries a value’s policy, the card reads Policy not updated and nothing is saved; an admin writes those in People properties.
  • She can’t change it from an automation, a Saved Agent run or an API token at all. An admin can also edit it directly, in the editor or from outside the app.

In Slack and Microsoft Teams she can’t change either policy, so a preference you state there isn’t written down. Tell her in the app, and the rule then applies to your Slack and Teams conversations too.

Ronja also reads two surfaces she wrote for herself: organization Knowledge and the Memory tab on your account page. She doesn’t add to either; new rules go in a policy. An admin corrects Knowledge with Curate in chat. Your personal memory she reads until you move it, below. See How Ronja learns and Curate Knowledge.

Lines from your personal memory offered as candidates appear above the editor on Your policy under Suggested from your personal memory: click one to add it as a rule, or Dismiss the lot. Nothing is added until you click.

While Ronja still reads your old memory, a Move your memory card sits at the bottom of the sidebar: “Ronja kept notes about how you work. Go through them once and keep what matters as your policy.” (With the sidebar collapsed it is a single icon.) Until you finish, Ronja keeps following your old memory exactly as before.

  1. Click Start. It opens a new chat that is just yours, with “Help me move my old memory into my policy.” already sent.
  2. Ronja reads your whole old memory and lists the few lines worth keeping as rules — at most eight, often fewer. She leaves out what happened in past conversations, numbers, facts about people and projects, guesses about you, and anything your policy already says, and tells you how many lines she left out. Then she asks: “Are these all relevant? Tell me which to drop or change, or anything I left out that you want kept.”
  3. Answer her: drop a line, reword one, or name one she missed. Nothing is written yet.
  4. Ronja asks for your approval with one card, Apply change and stop using your old memory, showing the lines that will be added to Your policy. Click Apply and stop using it. The lines are added after the rules you already have, and Ronja stops reading your old memory.

If nothing in your old memory is worth keeping, Ronja says so and offers to stop using it. Say yes and the card reads Stop using your old memory, with a Stop using it button; your policy does not change.

After you approve, the card in the chat says “Your old memory is no longer used.” Its Undo the policy change button takes the added lines back out of your policy — “Undo changes your policy only. Ronja still won’t use your old memory.”

Only a chat that is just yours can move your memory. A chat you have shared, a chat filed in a project, and a tangent of either cannot — and a request that came from a message someone else sent is refused, even in your own chat. Start a new chat from the card instead.

The same section is on Your policy while Ronja still reads the memory — Move your memory, above the editor, with the whole memory. Click Move it in a chat to open the same chat as Start, or copy the lines you want across by hand and click Done — stop using my old memory, then Stop using it to confirm. A very large memory shows only its newest 500 entries, and the section says so: “Only the newest 500 entries are shown; older ones aren’t.” Once you are done, the section moves below the editor.

A memory too long for a chat to read in full can only be moved by hand: the card then offers Open Your policy instead of Start, and the section has no Move it in a chat. The same goes for a User Read-Only member.

Click × on the card — “Remind me in a week” — to hide it for seven days. It then comes back, on any browser, until you finish.

Under Account → Memory, the tab says whether Ronja still reads your old memory and links to Move your memory. You can still edit a memory Ronja is reading; once you have finished there is nothing left to edit there.