Stream Chat

Start a streaming chat session with the Greenflash AI agent.

Requires a Growth or Enterprise plan.

The response is a Server-Sent Events (SSE) stream (Content-Type: text/event-stream). Each event follows the format:

event: <type>
data: <json>

Event types:

• tool_call — The agent is invoking a tool. Data: {"step": 1, "toolName": "...", "displayName": "..."}
• tool_result — A tool returned its result. Data: {"step": 1, "toolName": "...", "displayName": "..."}. For draftTicket and createTicket, the event also includes an output field with the tool's payload (see Ticket creation below).
• text_delta — A chunk of the agent's text response. Concatenate all deltas to build the full message. Data: {"text": "..."}
• done — The stream completed successfully. Data: {"conversationId": "...", "status": "complete", "usage": {"toolCalls": N, "tools": ["..."]}}
• error — An error occurred during processing. Data: {"error": "...", "code": "..."}

Multi-turn conversations: Pass previous messages in the messages array and reuse the conversationId returned in the done event.

Rate limits: This endpoint is rate-limited per tenant (requests/hour) and subject to token usage limits.

Ticket creation (two-step draft → confirm)

When the tenant has an active ticket-provider connection (e.g. Linear), the agent may emit a draftTicket tool call. The tool_result event includes an output field shaped like:

{
  "step": 2,
  "toolName": "draftTicket",
  "displayName": "Drafting ticket",
  "output": {
    "draft": {
      "provider": "linear",
      "title": "Billing page 500 for enterprise users",
      "description": "...",
      "target": { "teamId": "t_123", "teamName": "Core", "teamKey": "CORE" },
      "labelIds": ["lbl_bug"],
      "source": { "type": "conversation", "conversationId": "conv-abc-123" }
    },
    "availableLabels": [
      { "id": "lbl_bug", "name": "bug", "color": "#f00" }
    ],
    "dedupWarning": null
  }
}

API consumers should render this draft to the end user. To confirm (optionally with edits), send a follow-up user message asking the agent to call createTicket with the final payload. The createTicket tool_result event contains:

{
  "step": 3,
  "toolName": "createTicket",
  "displayName": "Creating ticket",
  "output": {
    "status": "created",
    "providerIdentifier": "LIN-99",
    "providerTicketUrl": "https://linear.app/acme/issue/LIN-99"
  }
}

status is "created" on success, "already_exists" when deduplication matched an existing ticket (the existing providerIdentifier / providerTicketUrl are returned), or the event may carry an error field (e.g. "provider_needs_setup").

event: tool_call
data: {"step":1,"toolName":"getConversations","displayName":"Searching conversations"}

event: tool_call
data: {"step":2,"toolName":"getUserRanking","displayName":"Ranking users by metrics"}

event: tool_result
data: {"step":1,"toolName":"getConversations","displayName":"Searching conversations"}

event: tool_result
data: {"step":2,"toolName":"getUserRanking","displayName":"Ranking users by metrics"}

event: text_delta
data: {"text":"Based on your"}

event: text_delta
data: {"text":" conversation data, the top complaints"}

event: text_delta
data: {"text":" from enterprise customers this week are..."}

event: done
data: {"conversationId":"conv-abc-123","status":"complete","usage":{"toolCalls":2,"tools":["getConversations","getUserRanking"]}}