Turn Agents into Action.
The agent action reads each row, follows your instructions, and writes its answer into the columns you choose. The agent can read one of your tables to choose from, or write new rows into another table.
Example 1 · Triage Support Message
Sort into your own categories.
The categories are rows in a Categories table, each with a description. The agent reads that table and picks the row that fits each message. The agent’s answer is a link to that row, so the two tables stay joined.
Example 1 · Triage Support Message: make the action. Add the inputs and outputs, write the instructions, pick the model, then save. The action reads from Categories.
Example 2 · Log Return Request: from the action page to Support Messages, where Mia’s message arrives, then to Returns, where the agent writes a row for each item
Example 2 · Log Return Request
Turn one message into several rows.
When one message holds more than one thing, the agent writes each thing as a row of another table. Mia asked to return two different items, so the agent added two rows to Returns.
Example 3 · Find Similar Products
Research, then write what the agent finds.
With web search on, the agent looks outside your data. When the Earl Grey Tin was added, the agent found five similar teas in UK shops and wrote one row for each, with the seller, price and link.
Example 3 · Find Similar Products: from the action page to Products, where the Earl Grey Tin arrives as a new row, then to Similar Products, where the agent writes a row for each listing it found on the web
Example 4 · Check Billing Message: from the Stripe connector to Support Messages, where four new messages arrive, are sorted as Billing, and get their Payment Check from Stripe
Example 4 · Check Billing Message
Check the facts in another system.
Connect an MCP server, and the agent can use the tools of that system. Here the agent reads the shop’s Stripe account. The agent finds Jon’s payment and writes what it found.
What you set on an agent.
These settings belong to the agent. Inputs, outputs, columns and conditions work the same as on every action: see Actions.
Instructions
Plain words: what to read, what to decide, what to write. The inputs go to the model as data next to the instructions, so you do not paste values into the text.
Model
Choose from OpenAI and Anthropic models, for each action. A smaller model costs less a run.
One table to read or write
Read: the agent can look up rows in one table of the workspace. Write: the agent adds rows to one table, with that table’s columns. No table: the agent works on its own row only.
Temperature
Set 0 for the same answer each time, for sorting and copying values out. Set the temperature higher for more varied writing.
Spend limit
The most one run can spend: 500, 2,000 or 6,000 credits, or as much as the model allows. You pay for what a run uses. A run that reaches the limit stops with an error.
Web search
Turn it on when the answer needs fresh facts from the web. Each run then takes more time and costs more.
Files
The agent reads PDF files and images itself. The agent reads Excel (.xlsx, .xlsm), CSV, Word (.docx), text and JSON files through tools. Not .xls, .doc, .pptx or .zip.
MCP servers
Give the agent the tools of another system, as in Example 4. The agent gets every tool of the server and acts there as the person who connected the server.
Good to know.
AI answers can be wrong.
Keep a person on the answers that matter. In Example 1, the reply is a draft that someone checks before it goes.
What an agent does in another system stays there.
morphcloud cannot undo what the agent did there. Connect with an account or key that can do only what the agent needs. Example 4 uses a Stripe test account.
Every change is on the row.
The row’s history shows each value the agent wrote, with its name and the time.
Every run is logged.
The log shows how long it took, the tokens sent and returned, and for a failed run, the error and where it came from.
A failed run does not repeat by itself.
Run the row’s actions again when you are ready. A failed run costs only the model calls that the run already made.
Add an agent to a table.
Start with a table, then tell the agent what to do with each row.
Security and compliance.
ISO 27001, SOC 2 Type 2 and GDPR are in progress. Oneleet shows the live status of each certification.
The agentic system builder
© 2026 morphcloud