Tailflow
Log inStart free
Skip to article
Browse documentation

Actions

Agent

Turn an event into a short explanation for a later action.

Set up the action

Connect Anthropic in Settings → Connections using the dialog’s model and credential controls. Select the connection on Agent and write a Prompt. This is your chosen model connection, not a separate Tailflow account setting.

The action receives the event and your prompt. Ask for a specific format and length. It does not automatically research wallet owners, browse websites, or verify facts beyond the supplied information.

Agent Action Configuration
Path 1
Every event
Agent
Selected action
1Agent
Continue on error
A failure stops later actions on this path.
Example. Write your instructions in Prompt. Later actions can use actions.agent_1.text; a test previews this request without generating an answer.Changes stay in this example. Nothing is saved or sent.

Write a useful prompt

Ask it to separate observed facts from interpretation and to say when data is missing. A model response may vary between runs and can be wrong; avoid wording that presents an unsupported inference as verified fact.

Agent Action Configuration
Path 1
Every event
Agent
Selected action
1Agent
Continue on error
A failure stops later actions on this path.
Example. Write your instructions in Prompt. Later actions can use actions.agent_1.text; a test previews this request without generating an answer.Changes stay in this example. Nothing is saved or sent.
Text
Summarize this transfer in two sentences using the amount, sender, and recipient in the event. Do not infer who owns an address. If a USD value is missing, omit the dollar estimate.

Use the answer

Place a notification after Agent and insert {{ actions.agent_1.text }}. The output also includes model. A model call takes time and uses 2 Tailflow credits when completed; usage of your connected model account may also apply.

Test run previews the request without calling the model. There is no generated answer for downstream templates in a dry run. Review the real response in Runs after a live event.

Slack Configuration with Agent Output
Path 1
1 · AgentSummarize the event
2 · SlackSend the answer
Slack
Example. Place Slack after Agent and select its text output. A dry test does not generate the Agent answer.Changes stay in this example. Nothing is saved or sent.