Chat

Chats are the individual conversations within an attempt — each chat corresponds to one scenario and holds the full message history between the learner and the AI persona.

Chat is a sub-namespace on the attempt artifact. Every operation is named chat_<op> and lives at POST /attempt/chat_<op>: chat_get, chat_message, chat_grade, chat_draft, etc. The CLI exposes them as glow attempts chat <op> for ergonomics; the wire path concatenates them under /attempt/.

What is a Chat?

A chat is a single conversational session between a learner and an AI persona within an Attempt. When a learner starts a simulation with three scenarios, three chats are created — one per scenario. Each chat holds the message thread, the persona configuration, attached documents, and any customization applied at runtime.

A chat contains:

• Chat Entry ID — unique identifier for this conversation
• Attempt ID — the parent attempt this chat belongs to
• Names / Descriptions — display copy
• Personas — the AI character(s) active in this chat (inherited from the scenario)
• Documents — reference materials available during the conversation
• Parameter Fields — resolved parameter values injected into persona instructions
• Scenarios — the scenario this chat is running
• Objectives — learning goals the chat is targeting
• Questions & Options — structured prompts for quiz-style interactions
• Images & Videos — media assets shown as context
• Problem Statements — the situation being addressed
• Flags — configuration flags for the chat session
• Departments — organizational groupings

Chats also support a draft system for customizing chat configurations before or during a session.

Sending a message

Quick Start

Get Chat Details via CLI

Calls below use $GLOW_INSTANCE_URL + $GLOW_TOKEN — see Authentication to export them once.

glow attempts chat get --body '{"chat_entry_id": "your-chat-entry-uuid"}'

Get Chat Details via API

curl -X POST $GLOW_INSTANCE_URL/attempt/chat_get \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $GLOW_TOKEN" \
  -d '{
    "chat_entry_id": "your-chat-entry-uuid",
    "attempt_id": "parent-attempt-uuid"
  }'

The response includes all hydrated resources: personas, documents, parameter fields, questions, options, and more.

Export a Chat

# CLI
glow attempts chat export --body '{"chat_entry_id": "..."}'
 
# API
curl -X POST $GLOW_INSTANCE_URL/attempt/chat_export \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $GLOW_TOKEN" \
  -d '{
    "chat_entry_id": "your-chat-entry-uuid",
    "attempt_id": "parent-attempt-uuid"
  }'

How Chats Connect to the Workflow

Chats are created automatically as part of step 5 when a learner starts an attempt:

You do not create chats directly via chat_create in normal flows — the attempt’s POST /attempt/start does it. chat_create is exposed for tooling that needs to seed chats outside the normal flow.

The chat sub-op surface

Every chat operation is a separate endpoint under /attempt/chat_<op>:

CLI dispatch is symmetric — glow attempts chat <op> (e.g. glow attempts chat message <chat_id> "...") maps 1:1 to POST /attempt/chat_<op>.

Grading a chat

Hints, feedback, and analyses

Live REPL (WS)

The same chat surface is also exposed over socket.io for live message flow. The CLI has an interactive REPL at:

# Open a socket.io REPL — type a line + Enter to send a chat_message
glow attempts chat live <chat_id>

Voice is handled separately by the chat_voice / chat_speak / chat_silence session ops (glow attempts chat voice|speak|silence, each a plain call to its POST /attempt/chat_* endpoint) — see the Voice pipeline for the full STT → LLM → TTS loop.

Searching and filtering chat resources

POST /attempt/chat_get supports search filters to narrow down the resources shown for a chat:

curl -X POST $GLOW_INSTANCE_URL/attempt/chat_get \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $GLOW_TOKEN" \
  -d '{
    "chat_entry_id": "your-chat-entry-uuid",
    "persona_search": "confused",
    "document_search": "FERPA",
    "persona_show_selected": true,
    "document_show_selected": false
  }'

Available search filters:

Working with drafts

Chat drafts let you customize a chat’s configuration — changing which personas, documents, parameters, and other resources are active.

# CLI
glow attempts chat draft --body '{
  "name": "Custom FERPA Chat",
  "description": "Modified chat with additional documents",
  "persona_ids": ["concerned-student-uuid"],
  "document_ids": ["ferpa-policy-uuid", "supplemental-guide-uuid"],
  "parameter_field_ids": ["location-param-uuid"]
}'
 
# API
curl -X POST $GLOW_INSTANCE_URL/attempt/chat_draft \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $GLOW_TOKEN" \
  -d '{
    "name": "Custom FERPA Chat",
    "description": "Modified chat with additional documents",
    "persona_ids": ["concerned-student-uuid"],
    "document_ids": ["ferpa-policy-uuid", "supplemental-guide-uuid"]
  }'

The draft endpoint supports optimistic concurrency through expected_version and returns the form_state with all selected resource IDs after save.

List your chat drafts:

# CLI
glow attempts chat drafts
 
# API
curl -X POST $GLOW_INSTANCE_URL/attempt/chat_drafts \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $GLOW_TOKEN"

Exporting a transcript

Common Operations

Related

• Attempt API Reference — every chat_* endpoint with full schemas
• Attempts CLI Reference — every glow attempts ... command
• Scenarios — the training situations that each chat runs
• Personas — the AI characters active in each chat
• Simulations — the training sessions that define which scenarios (and chats) are created