# Settings

> In short: The gear in the top bar opens Settings, where you can lower your spending limits for a production or for yourself, see why each limit is what it is, and see what changed it. You can only set a limit lower than the one the team running your workspace allows, and raising one asks you to confirm.

Applies to: v0.11 · Last reviewed: 2026-11-02 · Canonical: https://docs.zymcos.com/product/settings/

Open **Settings** with the gear in the top bar. It lists the limits and defaults that shape what
you can spend and what CODA may do for you, each with the value in force, **why** it is that value,
and where you can change it.

What you set can only be **lower** than what your operator allows. Your operator is the team that
runs your Zynth workspace: the figure they set is the ceiling, and a limit you set sits under it.
If they lower the ceiling below what you had set, the ceiling wins and the row says so: *Limited
by your operator to $15.00.*

## What you can change

<!-- settings-table:begin -->
| Setting | What it controls | Default | Who can change it |
|---|---|---|---|
| Daily spend cap for one production | The most the clicked generations of one production may cost in a rolling 24 hours, settled and reserved together. Runs are bounded by their own cap. | $25.00 | a member, for a production; your operator sets it for your whole workspace |
| Starting cap on a run's plan card | The spend cap a plan card starts with when your brief names none. You can change it on the card before you approve; it does not limit what you may approve. | $20.00 | a member, for a production; you, for yourself; your operator sets it for your whole workspace |
| Most one take may cost and be auto-approved | The limit a take is checked against once auto-approval is on. Auto-approval itself is off for everyone until it ships. | $0.50 | a member, for a production; you, for yourself; your operator sets it for your whole workspace |
| Counted for an unpriced image take | What an image take with no price counts at against the daily cap. | $0.25 | your operator only |
| Counted for an unpriced audio take | What an audio take with no price counts at against the daily cap. | $0.50 | your operator only |
| Counted for an unpriced video take | What a video take with no price counts at against the daily cap. | $5.00 | your operator only |
| Characters of draft CODA may write a day | How much script drafting your workspace may have written in a rolling 24 hours. Zero turns drafting off. | 300,000 characters | your operator only |
| Times a day CODA answers what you do on its cards | When you press a button on one of CODA's cards (a lock, a rights draft, an approval), CODA replies by itself with what happened and the next step. Each reply is a model call. Zero turns these replies off. | 50 replies | a member, for a production; you, for yourself |
| How far CODA goes before it stops for a click | Ask every time: every take is its own card. Plan first: CODA shows the whole plan and its total before the first spend, and a run pauses at every checkpoint. Autopilot within the cap: an approved plan runs to its cap and stops only where you chose to pause. A more careful scope always wins: your production can hold you at Plan first, and you can be more careful than your production. | Ask every time | a member, for a production; you, for yourself; your operator sets it for your whole workspace |
| Pause for my approval after the takes of an autopilot run | Where an autopilot run stops for you. Under Ask every time and Plan first a run pauses after every stage whatever this says; this only matters on Autopilot. | Pause for my approval | a member, for a production; you, for yourself; your operator sets it for your whole workspace |
| Auto-approve small takes | A priced single take under the limit above would skip its card. It is off for everyone and does nothing yet: it starts to when takes are priced by their length (R12), and turning it on will always need your approval. | Off | your operator only |
| Open items CODA is shown in every turn | How many of the production's open items (a scene with no shot, a card waiting, a failed take) CODA reads in every turn; the rest are counted. Your plan sets how many are useful, and you can lower it. | 20 items | a member, for a production; your operator sets it for your whole workspace |
| Plan items CODA is shown in every turn | How many open items of the production's plan CODA reads in every turn. | 12 items | a member, for a production; your operator sets it for your whole workspace |
| Lessons CODA is shown in every turn | How many of the lessons in force CODA reads in every turn, newest first. | 10 lessons | a member, for a production; your operator sets it for your whole workspace |
| Scenes CODA reads in detail | How many scenes around the one you have open CODA reads one by one; a longer script is a window of this many, so a feature costs what a commercial does. | 40 scenes | a member, for a production; your operator sets it for your whole workspace |
| Recent actions CODA is shown in every turn | How many of the latest actions in the production (by you, by CODA, by a collaborator or a reviewer) CODA reads in every turn; the rest are counted, and she can read further back when asked. Your plan sets how many are useful, and you can lower it. | 12 actions | a member, for a production; your operator sets it for your whole workspace |
| Check CODA's replies against the record | After CODA replies, a second, smaller model reads the reply against the production's record and can point at one sentence the record contradicts. It never changes anything or spends, and the correction in the chat is written by the app from the reply's own words and the record's, never by the model. It is on for every workspace and your plan sets how many replies are checked a day; only your operator can switch it off. | On | your operator only |
| Record a long conversation for CODA as it goes | When a conversation grows past what CODA reads at once, a record of it is made from your messages, the cards and what CODA did, and kept in view in place of the older messages. Off, CODA reads only its latest messages. A more careful scope wins. | On | a member, for a production; your operator sets it for your whole workspace |
| How you work | Which step the progress panel and CODA recommend next. It changes what is suggested and nothing else: you can do any step in any order, and it spends nothing. | Every scene's still before any clip | a member, for a production; you, for yourself; your operator sets it for your whole workspace |
| How much of a take you see and set | Simple: CODA chooses the model and the settings. Controls: you also see the model, length, shape and other common settings of a take, and its price as you change them. Expert: every setting the model offers, the seed, and why CODA would have chosen another model. It shows more of the same take and changes nothing about it. | Simple: CODA chooses | you, for yourself |
<!-- settings-table:end -->

A limit set for **a production** applies to everyone working in it. A limit set for **you** is
yours, wherever you work. When more than one applies, the **lowest** one is the one in force, and
the row says which it came from: *Set for this production*, *Set by you* or *Your operator's
default*. A limit you set to exactly your operator's figure still reads as set by you; your
operator's figure only decides how high it may go.

## Changing one

Press **Change** on a row. You can set a number for the production you have open, and, where the
row allows it, for yourself. Lowering a limit saves at once. **Raising** one, or going back to a
default that is higher than what is in force, asks first, in the word *raise*, because a higher
limit lets more be spent: *Raise this from $10.00 to $15.00 for this production?*

If someone else changed the same setting while you were looking, nothing is overwritten. The row
refreshes with what is there now, and you change it again from that.

Settings are changed by members who can curate the workspace. If you can only look, each row says
so and offers nothing to press.

## How far CODA goes before it stops for you

**Autonomy** has three steps, from most to least interruption:

- **Ask every time** (what everyone starts on): every take is its own card, and a plan, if you ask
  for one, pauses after each stage for your approval.
- **Plan first**: before the first spend of anything that needs more than one take, CODA shows the
  whole plan and its total and waits for your yes.
- **Autopilot within the cap** (Studio and Enterprise): an approved plan runs to its cap and stops
  only where you chose to pause. The pause after each stage of takes is on unless you say **Run
  straight through**.

A more careful choice always wins. If your production holds you at *Plan first*, you cannot be on
*Autopilot* there, and you can always be more careful than your production. Moving a choice toward
less asking says **Raise** and needs a member who can approve. These steps change only when CODA
stops for a click: they never raise your daily cap, never skip a check, and never let CODA lock,
delete or erase anything. **Auto-approve small takes** is off for everyone and does nothing yet; it
starts to when takes are priced by their length.

## What CODA may do

Below the limits, **What CODA may do** lists every kind of thing CODA can do, tool by tool, and the
models it may make takes with. For each you can choose:

- **Never**: CODA does not offer it, and is refused if it tries.
- **Ask me**: what CODA does today.
- **Without asking**: only for something free that you can undo, such as filing takes into a folder.
  It skips the confirmation card. Anything that costs money, and locking, deleting or erasing, never
  offers it: those always wait for your click.

Choose whether a choice applies to **this production** (everyone in it) or **just you**. Your
workspace's choice and this production's both count, and the most careful one wins. A model you rule
out is never used to make a take, whoever asks, and a [decision](https://docs.zymcos.com/product/decisions-and-charter/) that
names the models allowed rules out the rest, which the row says (*Ruled out by decision PDR-0004*).
On a card that asks to move files, **Don't ask again in this production** sets *Without asking* for
you in one click.

## What changed it

**History** on a row lists the recent changes, newest first, in words: *Raised from $10.00 to
$15.00 for this production*, *Went back to the default for you*. Each change is also recorded in
your workspace's audit record, with who made it and when, and only ever as a number.

## Asking CODA

Ask CODA what applies: *"What is my daily spend cap?"* CODA answers at once and says where each
limit comes from. Ask it to change one, *"Lower the daily cap for this production to $10"*, and
CODA puts a card in the conversation. **CODA never changes a setting itself.** Nothing changes
until you press the card's button, and a card that would raise a limit says so. Once you press it,
it is exactly the change Settings would have made.

## What is not here

Some figures are set only by your operator, and the page still shows them, with the reason:

- **What an unpriced take counts for.** A take on a model with no recorded price counts against
  the daily limit at a fixed estimate (see [Spending limits](https://docs.zymcos.com/product/spending-limits/)). A lower
  estimate would let more through, so it is not something a member can lower.
- **Your whole workspace's limits.** Where a row lists your workspace, your operator sets it for
  everyone; the row shows the value and says so.
- **Anything about how Zynth runs**: keys, addresses, rate limits, retention and which models
  are used. These are not settings, and the page does not pretend they are.
