A precision-first pull request reviewer that reports only the problems a change introduces, so no findings is a successful review. The fullest example of June's runtime: dependencies, parallel specialists and recovery in one workflow, with a benchmark that measures it.
junejune/train
A working CLI with deterministic tests, a precision-first benchmark and real-model runs.
A persistent writing agent in a person's own voice. It turns their edits into evidence, evaluates candidates on held-out cases, and changes only when they promote a candidate.
junejune/chatjune/copyjune/train
A working CLI with deterministic tests and a real-model run.
Building a workflow? Read the Code Review Agent: one run that plans, fans out to parallel specialists, verifies what they find and recovers from a failed branch.
Building a chat with feedback and evals? Read the Copy Agent: a conversation whose edits become evidence, and a new candidate that changes nothing until a person promotes it.
Just starting? Write your first workflow with the Quickstart, then come back.
Every guide has the same sections, in the same order, so you can read one application against another:
Problem: what the person using it needs, and what it does about that.
Architecture: the parts of the application and how data moves between them.
June modules used: which June imports it uses, and why.
Execution flow: the workflow or loop that does the main work, with excerpts from the source.
Using it: the commands a person runs, in the order they run them.
Testing: how it's tested without a model, and how it was checked with one.
What stayed in the app: the decisions that belong to the application, not to June.
Source and related docs: the code, and the docs pages for each module it uses.
A guide may add a section of its own between these. Code in a guide is an excerpt of a real file, linked to its lines, and the site's build fails if an excerpt stops matching the file.