Use Cases

BayLeaf AI Playground: concrete recipes for getting work done

These are illustrative scenarios drawn from real campus use, written for people who want to know what BayLeaf is good for before they invest time learning it. Each entry names the situation, the recommended approach (which service, which integrations, sometimes the literal prompt to start with), and any honest caveats.

These are not speculative. Each vignette below was lightly generalized from something a real BayLeaf user has actually done. Names, courses, and specifics have been blurred where appropriate, but the workflows, prompts, and integrations are the ones people are using right now.
Choosing between Chat and API: Reach for BayLeaf Chat when you want a quick, mobile-friendly browser session, possibly with built-in toolkits like web search, code sandbox, or Google Workspace. Reach for the BayLeaf API through a local coding-agent harness when you need the agent to read files on your own computer or when you want zero copies of the conversation stored on a BayLeaf service.

Faculty

Research: chasing down a half-heard funding opportunity

A colleague made passing verbal reference to a potentially relevant funding opportunity, and you want to follow up without disrupting the flow of the conversation, or losing the lead by tomorrow morning.

With BayLeaf: open BayLeaf Chat on your phone (you can install it as a Progressive Web App so it lives on your home screen like a native app). Start a new chat with the Basic model, enable the Web Context integration, and type something like:

I'm {your name}, faculty at UCSC. Help me track down the funding opportunity with a name like {vague phrase} and sketch how the work of my lab might fit it.

Send the message and put your phone away. By the time you return, the agent will likely have used multiple web searches in a row to figure out who you are and chase down the funder's details. The conversation syncs across devices, so you can continue from your laptop later.

Tip: If you often have research-flavored chats, put a line like User is {academic webpage URL} in your Chat system prompt so all future conversations are grounded in your academic identity without you having to reintroduce yourself each time.

Teaching: rescuing a misconfigured Canvas group assignment

Yesterday you posted a group assignment on Canvas. Today, reviewing submissions, you notice you forgot to mark it as a group assignment: now you have a disorganized pile of individual submissions. Students figured it out anyway (one person from each team submitted), but you still owe everyone group credit and need to identify the missing teams. Clicking through Canvas at scale is error-prone; pushing it back onto students invites confusion because most assumed a teammate had it covered.

With BayLeaf: start a chat with the Basic model and enable the Code Sandbox integration. If you've previously set up the canvaslms CLI in your sandbox, the agent can act quickly: create a new correctly-configured group assignment, append [deprecated] to the old one, and copy submissions and comments across, attaching each one to the right team. In the same conversation you can then ask "so, which teams still haven't submitted?" or "draft a Canvas announcement explaining the deprecation shuffle we just did."

Caveat: BayLeaf is not currently approved to handle the kinds of student data involved in this scenario. Use your judgment about whether the speed-and-accuracy tradeoff is appropriate for your context, and read the privacy notice first.

Students

Cross-checking a draft against a rubric without uploading the draft

You're writing a class assignment in Microsoft Word and want to cross-check your current draft against the rubric on Canvas. Your draft includes a personal story you're comfortable sharing with course staff but not with the BayLeaf administrator. Pasting the whole thing into a new Chat conversation would store a copy in BayLeaf's encrypted database. Is giving up that privacy the price of getting early feedback?

With BayLeaf: use the API instead of Chat. Pick a graphical coding-agent harness like Goose or OpenChamber (a CLI harness like OpenCode or Pi works too if you're comfortable in a terminal). Point the default agent at https://api.bayleaf.dev/docs/SKILL.md and it will walk itself through the setup. From then on, your agent reads the relevant files directly from your laptop, sends only what it needs to BayLeaf for inference, and keeps the only record of the conversation on your own machine. The personal story stays local.

Tip: This same pattern (local files, local transcript, BayLeaf only for inference) is the right default any time you're working with anything you wouldn't paste into a public chat: draft research, unpublished writing, anything covered by a confidentiality expectation.

Staff

Turning scattered email threads into a clean workflow doc

You're assembling workflow documentation for teammates in a Google Doc, but the only existing record of how the workflow has been done in the past lives in disorganized email threads and your own head. How do you gather that loose context and assemble it into something shareable?

With BayLeaf: use either the Google Workspace integration in BayLeaf Chat, or the gws CLI in a local agent powered by the BayLeaf API. Either way, the agent can search your email, follow links to other shared docs, and run a structured interview to grill you for missing details before writing a compact summary into either your working document or a private staging doc you review before publishing.

Tip: Ask the agent to produce the interview questions first, before doing any writing. You'll catch missing context and wrong assumptions earlier, and you'll end up with a better document for less total effort.

Contributing a use case

If you've used BayLeaf to do something concretely useful (and especially if it was something the obvious tools made painful), consider contributing your own write-up. Open a GitHub issue with the situation, the approach you took, and any caveats worth flagging, and we'll work it into this page.