Learning

Hefty gets smarter with every interaction. After each task completes, Hefty analyzes what happened and extracts reusable knowledge - automatically, in the background. You don't need to do anything special. Over time, this means Hefty becomes more effective, avoids past mistakes, and remembers your projects and workflows.

How It Works

After Hefty finishes a task, a learning cycle runs in the background:

Not every interaction produces new knowledge. Routine tasks (like reading a single file) are recognized as simple and skipped to avoid clutter. Learning only fires when there's something genuinely useful to remember.

Skills from Success

When Hefty successfully helps you with something non-trivial, it reviews what it did and may extract a reusable skill. For example, after helping you configure nginx with SSL, Hefty might save a skill called "nginx SSL setup" with notes on which commands to run, what config patterns to use, and which pitfalls to watch for.

Next time a similar task comes up, Hefty recalls that skill instead of starting from scratch - leading to faster, more reliable results.

Antipatterns from Failure

When an action fails or produces a bad result, Hefty doesn't just recover - it remembers what went wrong. For example, if a sed command corrupted a file because of unescaped characters, Hefty saves that as an antipattern with a note about what to do differently.

These antipatterns are recalled when Hefty faces similar situations, so it actively avoids repeating mistakes.

Entities from Context

Hefty picks up on notable names and concepts mentioned during your conversations - your teammates, the services you use, project-specific terminology. These are saved as entities. If you mention "the staging server runs on port 3001," Hefty can recall that fact later without you repeating it.

When relevant action results are produced (like API responses or configuration data), Hefty attaches them to the related entity as data artifacts - building up a rich reference that can be recalled on demand.

When the same person or concept comes up under a different name, Hefty recognizes it and merges the information together rather than creating duplicates.

Avoiding Duplicates

Before saving anything new, Hefty checks whether it already knows something similar. It uses meaning-based comparison, not just exact text matching - so a new skill about "Docker container deployment" won't create a duplicate if Hefty already has one about "deploying with Docker containers."

Connecting the Dots

After learning completes, Hefty links new knowledge together. A skill about "database migrations" might reference entities for your database tools. Entities accumulate data artifacts from action results. These connections help Hefty build a richer picture of your project over time, so the right context surfaces when you need it.