AI Workbook
Feature

Codelets. Deterministic logic for the parts of work that should not depend on prompts.

LLMs are excellent at reasoning support, synthesis, and interpretation. But many operational steps are best handled by deterministic logic. Codelets provide that layer: governed, sandboxed snippets of Python or JavaScript that execute in platform-managed runtime containers when workflows, tasks, agents, or other platform actions need precise control.

What codelets are

Codelets are user-defined and versioned. They sit in a registry, then run through dedicated execution endpoints.

That gives teams a dedicated extension layer for:

  • deterministic calculations and routing logic,
  • data cleanup and transformation before downstream steps,
  • task and webhook automation actions,
  • structured programmatic calls into platform operations and approved integrations.

The registry defines each codelet, while execution handles running it. Context and SDK support can be enabled for trusted cases where platform operations are required.

Codelets execute in a governed sandbox
The value is that custom logic runs in a controlled runtime with clear limits, not as an unbounded free-text prompt.

Why this matters

Why this matters
Separation of concerns
use LLMs for judgment, and codelets for control flow and calculation.
Predictable behavior
execution constraints are explicit (timeouts, memory, API budgets, network rules).
Repeatability
versioning gives reliable behavior over time.
Safer extension
policy controls determine what each codelet can do instead of open-ended scripting.
Built for the operational path
codelets are invoked from workflows, schedules, and webhook flows.
Next: Agents and AssistantsBack to Platform OverviewContact us