Agentic AI Automation via Sheet Assistant API

Find the median rate for 99214 office visit for each payer where state code is DC. Major payers only, exclude niche participants. Output requirements (strict): your last message should consist only of a pretty-printed json of results that is a list of objects, each object having keys "payer" and "median_rate". Do not include any other text in your final message.

Step 3: Poll until results available

Use this Sheet Assistant message history API call to poll for new messages from Sheet Assistant:

https://gigasheet.readme.io/reference/get_ai-handle-logs

While the agentic workflow is proceeding, Sheet Assistant may make many tool calls. Keep polling until you receive a message that has MessageType "assistant-response" containing your specified output.

Example output message with a result:

[
  {
    "Role": "assistant",
    "Content": "```json\n[\n  {\"payer\": \"UnitedHealthcare\", \"median_rate\": 177.720},\n  {\"payer\": \"Cigna\", \"median_rate\": 153.645},\n  {\"payer\": \"Aetna\", \"median_rate\": 108.280},\n  {\"payer\": \"Blue Cross Blue Shield - All Entities\", \"median_rate\": 87.810}\n]\n```",
    "MessageType": "assistant-response",
    "Timestamp": "2026-09-11T20:21:47.084863Z",
    "ClientStateId": ""
  },
  {
... previous messages, including tool call messages with MessageType "assistant-tool-output"
  }
]

It's recommended to poll approximately once every 15 seconds.

Conditions that are likely an error that you should escalate within your process:

• you have received an "assistant-response" message that does not conform to your output requirements, with no followup messages in the last 60 seconds (likely that Sheet Assistant is saying it failed and is no longer attempting your workflow)
• the total number of messages of type "assistant-tool-output" has exceeded 20 (likely that Sheet Assistant was unable to complete the workflow, consider revising your initial message or breaking the workflow into separate steps and running one step at a time)
• the most recent message is of type "assistant-tool-use" but with no new messages for over 180 seconds (query tool calls time out at 120 seconds, so anything significantly longer than 120 seconds after a tool call means something has failed and is not being retried)

Step 4: Extract the output

Extract the output from the final message use it in the next step of your process. For example, this is a Bash command with the jq utility to extract the answer from a file "response.json" that is the output of the API call in the previous step.

cat response.json | jq -r '.[0].Content' | sed '/^```/d' | jq .

On the above output, the command would produce this:

[
  {
    "payer": "UnitedHealthcare",
    "median_rate": 177.72
  },
  {
    "payer": "Cigna",
    "median_rate": 153.645
  },
  {
    "payer": "Aetna",
    "median_rate": 108.28
  },
  {
    "payer": "Blue Cross Blue Shield - All Entities",
    "median_rate": 87.81
  }
]

Make sure your instructions and your extraction code are aligned so that you can obtain the results. The jq extraction is just for illustration. Sheet Assistant is powered by a large language model, so it's recommended to implement a more robust extraction function to handle any formatting variations that may be generated in response to your initial instructions.