Chats & Tickets - ScaleDown

Email threads, Slack channels and support tickets. Four recipes, one per ScaleDown model.

Compress

Sits in front of your LLM. Open the full Python recipe.


Why it fits	Long email and Slack threads carry a lot of greetings, sign-offs, and quoted history; most of it is dead weight on the next call.
How it goes	Thread in → Compress strips the boilerplate, keeps the ask → the LLM works on the trimmed context.
Payoff	Fewer tokens into the drafting or answering call.

Replaces an LLM call. Open the full Python recipe.


Why it fits	People do not want to read a long thread top to bottom; they want the state of it.
How it goes	Thread in → Summarize produces the digest → surface it in the inbox, channel, or handoff.
Payoff	Thread digests without a frontier call each time.

Replaces an LLM call. Open the full Python recipe.


Why it fits	The asks are buried in prose; Extract turns them into fields a system can act on.
How it goes	Message in → tell Extract what to look for → it returns tasks, dates, owners → push to the tracker.
Payoff	Structured to-dos straight from the thread.

Replaces an LLM call. Open the full Python recipe.


Why it fits	Triage runs on every support message; volume makes per-message frontier cost add up fast.
How it goes	Message in → Classify assigns the label → route to the right queue, owner, or alert.
Payoff	Cheap, fast labeling at inbox volume.