examples: rag over your life in obsidian.md

examples/ts/rag-over-your-life-in-obsidian/conf.js

@@ -0,0 +1,7 @@
+const conf = () => {
+    const secrets = process.env.HOME + '/secrets.json'
+    console.log("fetching secrets.json from", secrets)
+    return require(secrets);
+}
+module.exports = conf;
+// ~/secrets.json: {"OPENAI_API_KEY": "sk-..."}

examples/ts/rag-over-your-life-in-obsidian/screenpipe.js

@@ -0,0 +1,64 @@
+const screenpipe = async (conf) => {
+    const openAIKey = conf.openai;
+    document.body.style.cursor = "wait";
+    const msg = window.getSelection().toString();
+
+    // Generate parameters for 3 different queries
+    const paramsResponse = await fetch('https://api.openai.com/v1/chat/completions', {
+      method: 'POST',
+      headers: {
+        'Content-Type': 'application/json',
+        'Authorization': `Bearer ${openAIKey}`,
+      },
+      body: JSON.stringify({
+        'response_format': { type: 'json_object' },
+        'messages': [
+          {
+            'role': 'user',
+            'content': `Based on this user selection: "${msg}", generate parameters as JSON for 3 different queries to screenpipe. 
+            Each query should have "q", "offset", and "limit" fields. 
+            Rules:
+            - q should be a single keyword that would properly find in the text found on the user screen some infomation that would help answering the user question.
+            Return a list of objects with the key "queries"`
+          },
+        ],
+        'model': 'gpt-4o',
+      })
+    }).then(res => res.json());
+
+    console.log(paramsResponse);
+    const queries = JSON.parse(paramsResponse.choices[0].message.content).queries;
+
+    // Query screenpipe 3 times with generated parameters
+    const screenpipeResults = await Promise.all(queries.map(async (query) => {
+      const response = await fetch(`http://localhost:3030/search?q=${encodeURIComponent(query.q)}&offset=${query.offset}&limit=${query.limit}`);
+      return response.json();
+    }));
+
+    // Ask ChatGPT to write an answer based on screenpipe results
+    const finalResponse = await fetch('https://api.openai.com/v1/chat/completions', {
+      method: 'POST',
+      headers: {
+        'Content-Type': 'application/json',
+        'Authorization': `Bearer ${openAIKey}`,
+      },
+      body: JSON.stringify({
+        'messages': [
+          { 'role': 'user', 'content': `Based on the user question "${msg}" and these data which corresponds to what has been seen on the user screen or through his mic: ${JSON.stringify(screenpipeResults)}, provide a comprehensive answer to the user intent.` },
+        ],
+        'model': 'gpt-4o',
+      })
+    }).then(res => res.json());
+
+    document.body.style.cursor = "default";
+    console.log(finalResponse);
+
+    if (!finalResponse.choices?.[0]?.message?.content) {
+      new Notice('Error from OpenAI');
+      new Notice(JSON.stringify(finalResponse));
+    }
+
+    return `${msg}${finalResponse.choices[0].message.content}`;
+  }
+
+  module.exports = screenpipe;

examples/ts/rag-over-your-life-in-obsidian/screenpipe.md

@@ -0,0 +1 @@
+<% tp.user.screenpipe(tp.user.conf()) %>