work learn
中文Get started

A learning layer for your AI workflow

Learn the English your work already needs.

Turn conversations with AI into a personal course you can actually reuse.

Open Work LearnConnect your agentSee how it works
saved from a real conversation
claudelearning layer on
Can we decouple the validation layer from persistence?
assistant

Yes. That keeps the domain rules independent from the database adapter and makes the module easier to test.

Worth learning01

decouple A from B

Separate two things so each can change independently.

your next use“We should decouple the API layer from the queue.”
The useful part stays close to the work.
Built for the way developers work
ClaudeChatGPTHermesOpenCodeterminal builds of agentsThe CLI captures agent sessions in your terminal, and clients that never get MCP support.Plus any agent that speaks MCP

Connect once. Learn from every agent.

The MCP server does the saving, searching, and reviewing. The Skill tells your agent when and how to use it. The CLI captures straight from your terminal.

01

MCP server

Connect it once to give every agent the same memory, search, and review.

save_material()
02

Skill

Drop in SKILL.md so your agent knows when to save and how to shape it.

organize this conversation
03

CLI

Run it in your terminal to capture agent sessions and clients without MCP support — or, on macOS, use the menu-bar Companion to capture anything you select with a hotkey.

learn capture

The same learning layer, everywhere

Your agents can have different voices. Your learning history should not.

your learningone memory
Sprimary
Skillguides the agent
Mserver
MCPsave and search
Cavailable
CLIterminal capture

A loop, not a notebook

The point is not more notes.

The point is the next time you can say it without thinking.

  1. 01

    Keep the good sentence

    Work Learn finds expressions worth reusing, not every word you happened to see.

  2. 02

    Make it yours

    See the natural version, why it works, and an example shaped around your codebase.

  3. 03

    Use it again

    Practice it in several formats — multiple choice, fill-in, scenario — and saved expressions resurface as a nudge exactly when you need them.

One place to actually use it

Your corpus, with the practice built in.

The web app is not another note dump. It shows what keeps coming up, then hands you a prompt to use it.

Your patterns

What keeps coming up

30d

Aggregated from your last 30 days of saved work.

12 materials8 questions23 expressions5 corrections

Topics

api 4migration 3debugging 2

Reusable expressions

  • → decouple the API from the queue
  • → take a closer look at the logs
  • → roll it out behind a flag

What to do next

Pick two high-frequency expressions and reuse them in today's work chat.

api · architecture

decouple the validation from the persistence layer

make the API to not couple with UI

Better
decouple the API from the UI

Why
"Decouple X from Y" is the fixed collocation.

What to do next

  1. reuse Use "decouple X from Y" in a new sentence about your work.
  2. recall Explain when you would say this.

Review, not a checkbox

01

take a closer look at the logs

I want to carefully examine the logs in detail

Each card hides the natural version until you ask. Recall it, grade how it went, and spaced repetition brings it back right before you would forget.

Start with the conversations you already have

Make your work teach you.

No new curriculum. No context switching. Just the useful English, when it matters.

What an account gives you

Signing up is what turns the endpoint above into something your agents can actually write to.

  • One corpus behind every agent. Save from whichever one you happen to be working in, review it all in one place.
  • Setup you do not have to work out. The app knows your client and hands you the config already filled in.
  • Tokens you control. Issue one per agent, revoke any of them the moment you want it to stop.
  • Yours only. Every item is filed under your user id, and the database refuses to hand it to anyone else.
  • Secrets never land. Every write path redacts API keys, tokens, and absolute paths before the text reaches the database.
Open Work LearnView SKILL.mdUsage guideFull setup docs
Remote MCP endpoint
https://work-learn.pages.dev/api/mcp

Twenty-one tools: create_session · save_material · search_corpus · get_review_items · mark_mastered · generate_practice · record_reuse · suggest_reuse · suggest_reuse_candidates · cluster_intents · and more

Auth: OAuth 2.1 with PKCE, or a personal access token you issue yourself.

Any client that speaks Streamable HTTP can use it. For the ones limited to local stdio, npx @work-learn/setup detects the client and writes the config for you.

Two ways to connect

Let your agent do it

  1. 1Sign up and open “Connect an agent”.
  2. 2It writes you a prompt with your endpoint and token already in it.
  3. 3Paste that into your agent. It finds its own config file and writes itself in.

Do it by hand

  1. 1Sign up and issue a token for the agent you want to connect.
  2. 2The app shows the config in the shape your client expects. Copy it across.
  3. 3Restart the client and ask it to save something worth keeping.

Both paths start with an account, because both need a token. The app walks you through whichever one you pick.

Sign in and connect