Lexicon Garden

chat.gel.assistant.defs

Repository @gel.chat did:plc:k2csuyvih44bjf2lex73x56v namespace authority
Details · defs
Kinddefs
Authorityassistant.gel.chat
Indexed
Compare Edit in builder → Fork →

#effort

string

The model's reasoning effort.

  • knownValues: none | low | medium | high | max

#encryptedContent

object

A sealed plainContent, or the sealed fields of the enclosing record. All user content is sealed, metadata stays in the clear.

Properties

blobs

array

Encrypted blobs the ciphertext refers to.

  • maxLength: 100
itemsblob

ciphertext

requiredbytes
  • maxLength: 1000000

keyId

requiredstring
  • maxLength: 128

nonce

requiredbytes
  • maxLength: 64

scheme

requiredstring
  • maxLength: 128

#extractedText

object

Cache for text pulled from a file.

Properties

method

requiredstring
  • knownValues: text | ocr

text

requiredblob
  • maxSize: 20000000
  • accept: text/plain

#filePart

object

An attached file. If the app can extract text, it will be stored in 'extracted' for the model.

Properties

extracted

ref
Reference

file

requiredblob
  • maxSize: 50000000
  • accept: */*

name

string
  • maxLength: 1024

#generationRequest

object

Properties

attempt

integer

The reply is written as '{messageKey}.r{attempt}'. Increase by 1 to ask for a regeneration.

  • minimum: 0
  • maximum: 1000
  • default: 0

effort

ref
Reference

model

ref

Model used in generating the response. If empty falls back to previous response model, then the user's default.

Reference

tools

array

The tools the assistant may use for this reply, by name, such as 'web_search' or 'mcp_<server>_<tool>'. Leave it out to use the tools the user has switched on in their settings.

  • maxLength: 64
itemsstring
  • maxLength: 128

#imagePart

object

An attached image.

Properties

alt

string
  • maxLength: 10000
  • maxGraphemes: 1000

image

requiredblob
  • maxSize: 20000000
  • accept: image/png, image/jpeg, image/webp, image/gif

#modelRef

object

The model that wrote a reply, or that a request asks for.

Properties

id

requiredstring
  • maxLength: 256

provider

requiredstring
  • maxLength: 128
  • knownValues: anthropic | openai | google | openrouter | cocore

#plainContent

object

The pieces of a message, in order: text, attachments, the assistant's reasoning, and tool calls with their results.

Properties

#providerData

string

JSON-encoded provider options for this part, such as reasoning signatures. Sent back unchanged when continuing with the same provider.

  • maxLength: 100000

#reasoningPart

object

The model's thinking or reasoning text.

Properties

text

string
  • maxLength: 500000

#sourcePart

object

URL source cited by the assistant.

Properties

title

string
  • maxLength: 3000
  • maxGraphemes: 300

url

requiredstring · uri

#textPart

object

Text written by the user or the assistant.

Properties

text

requiredstring
  • maxLength: 500000

#toolApprovalPart

object

The user's approval state for a tool call that's waiting in the reply this message is responding to.

Properties

approved

requiredboolean

callId

requiredstring
  • maxLength: 256

reason

string
  • maxLength: 3000

#toolCallPart

object

The assistant asked to run a tool. If the message has no result for this call it means the tool never ran.

Properties

approval

string

'pending' means the assistant stopped here and is waiting for user approval. When answering, write a user message with a toolApprovalPart for each waiting call. The reply to that message repeats each call, marked 'approved' or 'denied' and followed by its result, and replaces the waiting copy. Any waiting call the answer does not mention is treated as denied.

  • knownValues: pending | approved | denied

callId

requiredstring
  • maxLength: 256

input

requiredstring

JSON-encoded arguments.

  • maxLength: 200000

source

string

The tools source, such as the URL of an MCP server. Empty for the app's built-in tools.

  • maxLength: 1024

tool

requiredstring

The app's name for the tool, such as 'web_search' or 'mcp_<server>_<tool>'.

  • maxLength: 128

#toolResultPart

object

What a tool returned. It belongs to the call with the same callId in the same message.

Properties

callId

requiredstring
  • maxLength: 256

isError

boolean
  • default: false

output

requiredstring

Tool output as text or JSON-encoded data.

  • maxLength: 500000

outputBlob

blob

Used when the output is too long for the output record. Output will contain a summary in that case.

  • maxSize: 50000000
  • accept: */*

#usage

object

Token counts for the reply.

Properties

inputTokens

integer
  • minimum: 0

outputTokens

integer
  • minimum: 0

reasoningTokens

integer
  • minimum: 0

Similar lexicons

semantic match

Lexicons whose schemas describe something close to this one. Check them before publishing a new one that overlaps.

  • network.sharedcomputer.chat.defs

    The model's reasoning effort.

  • chat.gel.assistant.message

    record

    One message in a chat, from the user or the assistant. A reply's key is the user message's key plus '.r' and the attempt number, such as '.r0' or '.r1'. The app creates the reply with status 'pending' and no content, it fills it in when generation ends. Creating fails if the key is taken, so each attempt gets exactly one reply. Sort replies by attempt number rather than by key, since '.r10' sorts before '.r2'.

  • network.sharedcomputer.chat.message

    record

    One turn in a conversation. A reply uses its parent's key plus '.r' and the generation attempt, e.g. '.r0', '.r1'.