Skip to main content

Agents

Abstra Agents are now used as a function call, not as a workflow stage.

Use run_agent from abstra.ai inside a regular Python stage (typically a Tasklet).

from abstra.ai import run_agent
from abstra.tasks import get_trigger_task

task = get_trigger_task()

result = run_agent(
    prompt=f"""
You are a customer support agent.
Customer: {task.payload.get("customer_name")}
Request: {task.payload.get("request")}

Classify this request and propose a response.
""",
    max_steps=30,
)

print(result)

How It Works Now​

  1. A stage script calls run_agent(...).
  2. The prompt is sent to the LLM runtime.
  3. The agent runs a reasoning loop and can call any tool you expose.
  4. The run ends when it finishes or reaches max_steps.

Migration Notes​

  • Old Agent Stage: removed.
  • Old prompt-template markdown (.md + Jinja2): replace with Python strings (usually f-strings).
  • Old permission picker: replace with tools=[...] passed to run_agent.

See:

When to Use run_agent​

Use it when the process requires non-deterministic reasoning and tool selection, such as:

  • Working with unstructured content (emails, documents, screenshots)
  • Multi-step tasks where each next action depends on previous results
  • Context-dependent routing/triage decisions

For fixed logic with deterministic steps, a regular Tasklet script is usually simpler.