Continue-As-New

Continue-As-New resets the workflow's event history while carrying forward its state, enabling indefinitely long-running workflows without hitting history size limits.

When to use

Long-running workflows that iterate indefinitely
When workflow history approaches size limits (~50K events or 50MB)
Workflows that need to run for weeks or months

How it works

When continue_as_new() is called, the current run completes and a new run starts fresh from the beginning with the provided state. All accumulated event history is discarded.

Important: Your workflow's run method must accept a state parameter to restore its state when continuing as new.

import mistralai.workflows as workflows

@workflows.workflow.define(name="long-running-processor")
class LongRunningProcessor:
    @workflows.workflow.entrypoint
    async def run(self, page: int = 0, total_processed: int = 0):  # State parameters are required
        # Process current page
        result = await process_page(page)
        total_processed += result.count

        # Check if history is getting large
        if workflows.workflow.should_continue_as_new():
            # Continue with fresh history, passing current state
            workflows.workflow.continue_as_new({"page": page + 1, "total_processed": total_processed})

        return await self.run(page + 1, total_processed)

import mistralai.workflows as workflows

@workflows.workflow.define(name="long-running-processor")
class LongRunningProcessor:
    @workflows.workflow.entrypoint
    async def run(self, page: int = 0, total_processed: int = 0):  # State parameters are required
        # Process current page
        result = await process_page(page)
        total_processed += result.count

        # Check if history is getting large
        if workflows.workflow.should_continue_as_new():
            # Continue with fresh history, passing current state
            workflows.workflow.continue_as_new({"page": page + 1, "total_processed": total_processed})

        return await self.run(page + 1, total_processed)

Checking history size

Use should_continue_as_new() to check whether it's time to reset:

if workflows.workflow.should_continue_as_new():
    workflows.workflow.continue_as_new({"page": next_page, "count": total})

if workflows.workflow.should_continue_as_new():
    workflows.workflow.continue_as_new({"page": next_page, "count": total})

This returns True when the event history is approaching the system limit.