Admin ConsoleAdminister

Platform settings

Register models and providers, set prices and context windows, and set rate limits and retry behaviour for your organization.

Required role

Owner or Admin of the active organization.

How settings work

Open Settings from the Administration group (the top bar calls it Platform Settings). Every setting starts at a built-in default. When you change one, your value is saved for the active organization only and marked overridden; everything you haven't changed keeps following the built-in default. Changes take effect within about a minute.

The page shows one card per category, with how many settings it holds and how many are overridden. Select a card to open it and All settings to go back. The cards are General, Model Settings, PDF Citations, Rate Limits, Providers, Logging and Prompt Management (see Prompt management).

Change a single setting

1
Open the card
Each setting shows its name, description and Default.
2
Enter a value
On/Off settings use a menu; others a text or number box. A Save button appears once you change the value.
3
Select Save
The setting is marked overridden. To go back to the default, select the reset icon next to it.

Settings reference

General

Setting	Default	Meaning
Default Provider	openai	Provider used when a request doesn't name one.
Default Timeout (s)	300	How long a single model call may take before it's retried.
Default Max Retries	3	How many times a transient failure is retried before the request fails.

Rate Limits

Setting	Default	Meaning
Model RL — Enabled	On	Turns on the per-model request caps set in the RPM (model cap) column of Model Settings.
Max Rate-Limit Sleep (s)	300	How long a request waits for a rate limit to clear before failing with 429.
User RL — Enabled	Off	Caps every user independently with the four limits below, on top of their tier.
User RL — per 1 min, per 1 hour, per 6 hours, per 1 day	0	Requests per user per window. 0 means unlimited.

Per-person limits by model and budget are set on tiers instead; see Tiers and limits.

PDF Citations

Setting	Default	Meaning
PDF Citations — Enabled	Off	Lets the agent add clickable references to answers drawn from an attached PDF, showing the document and printed page. This is the organization-wide switch; each person must also turn it on for themselves.

Logging

Setting	Default	Meaning
Enable Request Logging	On	Records requests for usage analytics. Turn it off and usage analytics stop recording.

Model Settings

The model registry: one row per model. A model must be here, with a provider, before it can be allowed on a tier or picked as a judge or agent model. Rows you haven't changed follow the built-in catalogue.

Column	Meaning
Model	The model's name.
Served as	The provider the model is published under. Clients see it as `<provider>/<model>`, and tier allow-lists match that exact string.
Configure	Per-provider details: the ID each provider knows the model by, input, output, cache-read and cache-write prices (dollars per million tokens), the context window and the maximum output tokens. Also lists other providers the model can be routed to.
Tools	The model supports tool calls.
Allowed tools	Needs Tools on. Empty allows every tool; otherwise the model may use only the ticked LatentCode tools and other tool requests are rejected.
Vision	Accepts images.
PDF	Accepts PDF attachments. Turn on only for models whose provider accepts PDFs, or requests with a PDF fail.
Reasoning	The model can reason. The second Reasoning column declares which effort levels it accepts and whether reasoning can be turned off. Declare only what the model really supports: an undeclared level fails the request.
JSON mode	Supports JSON output mode.
Stream	Supports streaming.
RPM (model cap)	Requests per minute allowed to this model, enforced when Model RL — Enabled is on. Required; new rows start at 100.

Add a model

1
Select Add
In the Models section of Model Settings.
2
Fill in the minimum
A Model name, Served as, and the ID the serving provider uses for it. Prices, limits and capabilities are pre-filled for known models. For a model that isn't known on its provider, enter both the input and output price (0 is allowed), the context window and the max output tokens yourself: Save changes refuses the row without them.
3
Check and correct
Open Configure… to see what each provider resolves to and correct anything wrong. Test live calls the model at each reasoning level it declares, using the row as you've edited it, including unsaved prices. It spends tokens. If the row names several providers, pick the one to test.
4
Select Add row, then Save changes
Edits to this section aren't saved until you select Save changes at the bottom. Discard throws them away.

Section tools

Export downloads the section as a JSON file. Provider API keys are exported masked.
Import loads an exported file for the same section and shows which rows it would add, change and remove. Untick any change you don't want, then select Save changes.
The row reset icon reverts one row to its built-in default. The trash icon at the top of the section resets every override in it.
The filter box narrows the rows shown.

Providers

The endpoints and credentials the gateway uses to reach model providers. One row per provider.

Column	Meaning
Name	The provider name used in Served as and in model IDs.
SDK Provider	The vendor this endpoint belongs to. Set it only if you gave the provider a custom name.
Protocol	Leave blank unless your provider needs a specific API.
Base URL	The endpoint URL.
API key	The provider credential. After saving it's shown masked.
Extra headers	JSON object of headers sent with every call.
Extra config	JSON object of provider-specific options.

Secrets in exports

API key is masked in exports, but anything you store in Extra headers or Extra config is exported as written. Treat exported provider files as secret.

Deleting a provider that models are served from leaves those models with no route; the page warns you and lists them.

Troubleshooting

A model doesn't appear in the tier editor

Only models with a provider are offered. Check its Served as column and that the provider exists.

Save changes is disabled with "served as a host with no id"

A model is served by a provider but has no ID for it. Open Configure… and add the ID.

Save changes fails with "`input` and `output` rates are required together" or "`context_window` is required"

The model isn't known on the provider serving it, so its prices and limits can't be filled in for you. Open Configure…, enter both prices (0 if it's genuinely free), the context window and the max output tokens for that provider, then save.

Requests fail with "… has no price, so every call would bill at zero"

The model has no price on the provider serving it, so the gateway refuses to call it. This also applies to rows saved before prices were required. Set both the input and output price for that provider under Configure… and select Save changes.

Requests fail after adding a model

Check the model ID the provider expects, the provider's API key, and the Reasoning declaration. An effort level the model doesn't support fails the request.

My change hasn't taken effect

Changes take effect within about a minute. Select Refresh to confirm the saved value.