Send A2A tasks

Agent2Agent (A2A) suits agents that send a structured task and collect the result later, rather than calling tools one by one.

Setting	Value
Agent card	GET https://api.inside.chat/.well-known/agent-card.json, no authentication
Endpoint	POST https://api.inside.chat/a2a
Headers	Authorization: Bearer ..., Content-Type: application/json, A2A-Version: 1.0
Token	An installation token, or an OAuth token requested with resource=https://api.inside.chat/a2a. REST and MCP OAuth tokens are rejected.
Methods	SendMessage, GetTask, ListTasks, CancelTask

Send a task and poll

1. Send the task with SendMessage. The message has exactly one data part with a skill and its arguments.

   curl -s https://api.inside.chat/a2a \
     -H "Authorization: Bearer $INSIDE_GATEWAY_TOKEN" \
     -H "Content-Type: application/json" \
     -H "A2A-Version: 1.0" \
     -d '{
       "jsonrpc": "2.0",
       "id": 1,
       "method": "SendMessage",
       "params": {
         "message": {
           "messageId": "weekly-insights-2026-10-13",
           "role": "ROLE_USER",
           "parts": [{
             "data": {"skill": "support_insights", "arguments": {"website_id": "web_4927_123", "range": "last-7-days"}},
             "mediaType": "application/json"
           }]
         },
         "configuration": {"returnImmediately": true, "acceptedOutputModes": ["application/json"]}
       }
     }'

2. Poll with GetTask until the task reaches TASK_STATE_COMPLETED, TASK_STATE_FAILED or TASK_STATE_CANCELED.

   curl -s https://api.inside.chat/a2a \
     -H "Authorization: Bearer $INSIDE_GATEWAY_TOKEN" \
     -H "Content-Type: application/json" \
     -H "A2A-Version: 1.0" \
     -d '{"jsonrpc":"2.0","id":2,"method":"GetTask","params":{"id":"task_..."}}'

3. Read the artifact. A completed task has one JSON artifact of at most 128 KiB.

With returnImmediately: true the task returns as TASK_STATE_SUBMITTED or TASK_STATE_WORKING. Without it, the request waits for the final state.

Skills

Free text isn’t interpreted. Each skill takes structured arguments:

Skill	Arguments	Required scopes
support_insights	website_id, range (today, last-7-days, last-month or date), and start, end, timezone for date	reports_read, conversation_read
conversation_evidence	conversation_id, optional limit (1 to 100, default 50) and cursor	conversation_read, message_content_read, contact_read_pii, plus the message_content and pii flags
configuration_inspection	website_id, path (/ai/features or /ai/features/{feature_id}), optional limit and cursor	config_read
contact_inspection	contact_id	contact_read; personal fields need contact_read_pii and pii
offline_language_detection	feature_id, expected_version, context: {"type":"text","text":"...","default_language":"en"}	ai_features_read, ai_execute

Task rules

• Retries. If a reply is lost, resend the same messageId, skill and arguments. You get the same task back without running it again.
• Paging. Transcript and configuration results are single pages marked partial: true. Request the next page with a new messageId.
• Lifetime. Tasks expire after 15 minutes, or sooner if the credential expires. Only the credential that created a task can read it; a new OAuth token can’t see tasks created by an earlier one.
• Cancelling. CancelTask stops a result from being published. Work already sent to the website may still finish.
• Capacity. Each installation can hold up to 100 unexpired tasks, with up to 5 running at once. Every A2A request uses one unit of the app’s shared quota, and a new task also uses one for the underlying read.
• Streaming and push notifications aren’t supported.