diff --git a/README.md b/README.md
index c37ead4..caeeedf 100644
--- a/README.md
+++ b/README.md
@@ -1,43 +1,43 @@
 # Chat-DBT
 
-## MVP
+## Usage
 
--   [ ] Implement the options into the web interface
-    -   [ ] confirm
-    -   [ ] automatic corrections
-    -   [ ] keep-context
--   [ ] Use `inquirer` instead of `prompts` (has an editor prompt, a yes/no/other prompt, and seems easier to tweak to implement history)
--   [ ] Arrow up/down to get the previous queries
-    -   [ ] CLI
-    -   [ ] Web UI: https://mantine.dev/core/autocomplete/
--   [ ] Editable SQL query before confirmation (and keep the modifications in the history)
--   [ ] Editable SQL query after an error (and keep the modifications in the history)
--   [ ] Changesets + publish
--   [ ] Complete this readme
+```sh
+npx chat-dbt [options]
+npx chat-dbt --help
+```
+
+### Command-line interface
+
+```sh
+npx chat-dbt --database postgres://user:password@localhost:5432/postgres --key [your-openai-key] --org [your-openai-org]
+```
 
-## Post-release
+### Web interface
 
-Library
-Stats eg time per request, tokens used, etc
+```sh
+npx chat-dbt web --database postgres://user:password@localhost:5432/postgres --key [your-openai-key] --org [your-openai-org]
+```
 
--   [ ] Web interface in dark/light mode
--   [ ] Add CI tests for both the CLI and the web interface
--   [ ] Nice to have (but complicated): when the context is preserved, allow queries from previous data e.g. "translate the previous result into french"
--   [ ] `--hide-sql`
--   [ ] `--hide-result`
--   [ ] Input from stdin/output to stdout (combined with the above)
--   [ ] Output sql (on success) to file
--   [ ] Output result to file
--   [ ] Verbose level, timestamps etc
--   [ ] Comment the code
+### Use a `.env` file
+
+<!-- TODO -->
+
+### Use another OpenAI model
 
-## Done
+<!-- TODO -->
 
--   [x] Change the terminology "retries" to "ask-corrections"
--   [x] Add a `--confirm` option to prompt the user before executing the query
--   [x] Add a `--model` option to choose the OpenAI model
--   [x] Spinner
--   [x] Manual requests for corrections
+### Ask for confirmation before executing the SQL query
+
+<!-- TODO -->
+
+### Keep context between queries
+
+<!-- TODO -->
+
+### Change the format of the result
+
+<!-- TODO -->
 
 ## Development
 
@@ -66,4 +66,34 @@ pnpm run dev:web
 
 ### Use the CLI
 
-TODO nodemon
+<!-- TODO nodemon -->
+
+## Roadmap
+
+-   [ ] Implement the options into the web interface
+    -   [ ] confirm
+    -   [ ] automatic corrections
+    -   [ ] keep-context
+-   [ ] Use `inquirer` instead of `prompts` (has an editor prompt, a yes/no/other prompt, and seems easier to tweak to implement history)
+-   [ ] Arrow up/down to get the previous queries
+    -   [ ] CLI
+    -   [ ] Web UI: https://mantine.dev/core/autocomplete/
+-   [ ] Editable SQL query before confirmation (and keep the modifications in the history)
+-   [ ] Editable SQL query after an error (and keep the modifications in the history)
+-   [ ] Changesets + publish
+-   [ ] Complete this readme
+
+### Post-release
+
+-   [ ] Use the npm package as a library
+-   [ ] Statistics e.g. time per request, tokens used, etc
+-   [ ] Web interface in dark/light mode
+-   [ ] Add CI tests for both the CLI and the web interface
+-   [ ] Nice to have (but complicated): when the context is preserved, allow queries from previous data e.g. "translate the previous result into french"
+-   [ ] `--hide-sql`
+-   [ ] `--hide-result`
+-   [ ] Input from stdin/output to stdout (combined with the above)
+-   [ ] Output sql (on success) to file
+-   [ ] Output result to file
+-   [ ] Verbose level, timestamps etc
+-   [ ] Comment the code
diff --git a/shared/README.md b/shared/README.md
new file mode 100644
index 0000000..b785c03
--- /dev/null
+++ b/shared/README.md
@@ -0,0 +1,3 @@
+# Shared code
+
+Code used by both the CLI and the Web interface