Skip to main content
PUT
Update institution settings

Authorizations

x-access-token
string
header
required

JWT token passed in x-access-token header

Body

application/json

Any combination of authorized institution properties

conversationModel
string

Default conversation model for the institution

conversationFallbackModel
string

Fallback conversation model — answers when the conversation model fails before any reply started. Requires the institutions.edit entitlement. '' inherits the platform default (System Models → Conversation Fallback, unset until an operator picks one); otherwise a catalog conversation model or a custom:<aimodelId> reference to one of this twin's active custom conversation models. Any other value is rejected with 400.

maxCompletionTokens
integer

Maximum completion tokens for AI responses

reasoningEffort
string

Reasoning effort level (none, low, medium, high, max)

extendedContext
boolean

Whether extended context is enabled

about
string

About text for the institution

personaMode
boolean

Persona twin toggle (admin-gated — requires the institutions.edit entitlement, silently stripped otherwise). When true the platform prompt becomes persona-aware — capability and onboarding questions are answered in the twin's own voice, without tool lists or platform guides.

twinWorkflowManagement
boolean

Twin workflows switch (admin-gated — requires the institutions.edit entitlement, silently stripped otherwise; not super-only). Default true: the twin's admins can ask it in web chat to create, change, run and schedule read-only workflows (also needs the TWIN_WORKFLOW_TOOLS server flag). false removes the chat tools; existing workflows keep running.

personalisationAsked
boolean

Whether personalisation has been asked

disableToolUseInstructions
boolean

Whether to disable tool use instructions

kmeanScore
number

K-means score threshold for RAG

ragLimitChunks
integer

Maximum number of RAG chunks to retrieve

toolsDisabled
boolean

Whether tools are disabled for AI conversations

roleEfforts
object

Per-subagent role reasoning preferences, authorized like roleModels (institutions.edit). Keys are the role slots: conductorModel, plannerModel, designerModel, coderModel, testerModel, writerModel, researcherModel, agenticLoopModel, preflightModel, specWriterModel, planReviewerModel, judgeModel, vision. Whole-map update; preserve sibling slots. Empty clears a slot. Provider capability limits still apply; Astra None maps to Low. xhigh is sent verbatim to Anthropic adaptive-thinking models and as max elsewhere. An unset coder effort runs Claude Opus 5.5 at xhigh.

tierModels
object

Per-mode model pins (Map mode → role slot → model id), authorized like roleModels (institutions.edit). Modes: basic, advanced, expert, mythical (auto has no pool). Slots: the roleModels slots plus the coder level slots coderUpgradeModel (harder tasks) and coderVisualModel (games and simulations), which are accepted ONLY here, never in roleModels. A mode pin outranks the twin-wide roleModels pin and the platform per-mode pick; '' or null clears one pin, a null mode clears that mode. A coder level pin is used when the run's coder dispatch picks that level; with a twin-wide or mode coder pin and no level pins, the automatic level choice stays off. Values must be model-id slugs the catalog allows for that role (400 otherwise).

tierEfforts
object

Per-mode reasoning effort (Map mode → role slot → effort), same vocabulary, authorization and whole-map semantics as roleEfforts. Modes: basic, advanced, expert, mythical. Slots: the roleEfforts slots except conductorModel (the conductor has no per-mode effort), plus coderUpgradeModel and coderVisualModel (the effort for a coder dispatch that ran that level). Order on a run: this mode cell, then roleEfforts, then the platform effort, else the model's own default. Empty clears a cell.

enableModeration
boolean

Route incoming prompts through the moderation model before they reach the main LLM

alwaysCiteSources
boolean

Add an explicit citation directive to the system prompt so the model footnotes RAG / web sources in its answers

toolResultsMaxChars
integer

Character cap for individual tool / scrape / vault results before they are trimmed and saved to the IP Vault. Default 60000.

Required range: 1000 <= x <= 200000
toolCallBudget
integer | null

Maximum tool calls a single answer may make before the Digital Twin must write its answer. Unset (null) uses the platform default (20); a Digital Twin can only lower the platform cap.

Required range: 1 <= x <= 50
rtEnabled
boolean

Enable Convo / Speech-to-Speech (realtime voice) mode for this instance

rtAdminOnly
boolean

Restrict Convo mode to admin users only

rtTextInputEnabled
boolean

Allow text input alongside speech while in Convo mode

rtModel
string

System model id for Convo / Speech-to-Speech (provider auto-derived from the model id)

rtVoice
string

OpenAI Realtime voice (alloy, ash, ballad, coral, echo, sage, shimmer, verse, cedar, marin)

gemini_rt_voice
string

Gemini Live voice (Puck, Charon, Kore, etc.)

xai_rt_voice
string

xAI realtime voice (eve, ara, rex, sal, leo)

openai_live_voice
string

OpenAI GPT-Live voice (Quartz, Ripple, Vesper, …). Empty = each member picks their own.

rtVADEagerness
enum<string>

OpenAI realtime voice-activity-detection eagerness

Available options:
low,
medium,
high
gemini_rt_start_sensitivity
enum<string>

Gemini Live interruption sensitivity — how readily the twin stops talking when a member speaks over it. low (default) ignores background noise but is harder to interrupt; high is Gemini's own default and easy to talk over.

Available options:
low,
high
rtNoiseReduction
enum<string>

OpenAI realtime noise reduction profile

Available options:
,
near_field,
far_field
rtTranscriptionLanguage
string

ISO 639-1 language hint for OpenAI realtime transcription (blank = auto-detect)

disableClipboardForUser
boolean

Hide clipboard controls from non-admin users in this institution

locationEnabled
boolean

Allow this institution's users to opt-in to sharing browser geolocation with the assistant for location-aware answers. Defaults to true.

guestUI
boolean

Strip the UI down to a chat-only minimum surface for non-admin users

theme
enum<string>

Forced UI theme for all users in this institution. '' = user directed, 'dark' = force dark mode, 'light' = force light mode

Available options:
,
dark,
light
creditCaps
object

Per-user quota configuration. Settable by admins/managers with the institutions.edit entitlement when the account has not enabled caps; ignored (read-only inherited) when the account owns caps.

Response

Institution updated successfully

success
boolean
Example:

true

message
string
Example:

"Institution updated!"