Plans

Plans let you automate multi-step workflows that run in the background. Each plan consists of sequential steps that execute independently, allowing you to streamline repetitive tasks and complex processes.

Plans are specifically designed to handle tasks with a very wide context window—such as analyzing extensive document libraries or processing large datasets—that might be too resource-intensive for a standard chat interaction. Furthermore, Plans enable you to execute these workflows on a recurring schedule. This makes them ideal for continuous monitoring tasks, such as tracking industry news feeds every week or performing periodic data audits, ensuring that long-form automation remains reliable and efficient without interrupting your immediate workflow.

How plans work

Plans are scoped to both a specific user and a specific assistant. When you create a plan, it becomes available for that assistant and can be started from any conversation within that assistant.

Plan structure

A plan consists of three core components:

• Steps: Individual tasks that execute one after another.
• Step configuration: Internal settings for each step (action type and expected inputs).
• Schedule (optional): Automated recurring execution.

Working with artifacts

Plans are highly effective at generating and refining content across multiple steps. They can create and update artifacts (documents or files) throughout an execution. If step 1 creates an artifact, step 2 can fetch that artifact and build upon it. This enables collaborative, iterative document creation within a single execution.

Beta status and current limitations

Plans is currently a Beta feature. To request access and begin using Plans, please contact Intric support.

We are actively developing this functionality and iterating based on direct feedback from our beta users. Because we are in a period of rapid development, the limitations listed below are subject to change and will likely evolve quickly as we continue to iterate on the platform:

• Sequential only: Plans execute one step at a time; parallel processing is not supported.
• No partial restarts: You cannot restart a failed plan from a specific step; you must rerun the entire plan.
• No chat continuation: Once a plan completes, the interaction for that specific run is finished.
• File uploads: The frontend currently requires files to be uploaded per-step rather than for the whole plan at once.
• Artifact editing: You cannot directly edit artifacts created by plan executions or continue working on them in a standard conversation. If you are not satisfied with the result, you must adjust the plan instructions and rerun the execution.

Best practices for success

• Write clear instructions: Since steps run in isolation, be explicit about what each step needs to do. Don’t assume the AI remembers context from the “conversation” unless it is explicitly in the step prompt.
• Leverage expected inputs: Use this feature to ensure you never start a plan without the necessary files.
• Test manually: Always run a plan manually to verify the output before setting up an automated schedule.
• Use appropriate action types: Although automated, thinking in terms of “Write,” “Search,” or “Analyze” helps you structure better steps.