@vierratale/ai 0.1.0-beta.14 → 0.1.0-beta.16

package/src/providers/cortex.js

@@ -2,25 +2,6 @@ import { BaseProvider } from './base.js';
 import { Catalog } from '../catalog.js';
 import { Config } from '../config.js';
 
-// Keep the prompt a local engine must re-process per call bounded: the
-// system prompt plus the most recent messages. Sending the whole growing
-// history makes prompt-processing time unbounded on slow hardware.
-const MAX_CONTEXT_MESSAGES = 12;
-
-function buildEngineMessages(systemPrompt, messages) {
-  const engineMessages = [];
-  if (systemPrompt) {
-    engineMessages.push({ role: 'system', content: systemPrompt });
-  }
-  for (const msg of messages) {
-    engineMessages.push({ role: msg.role, content: msg.content });
-  }
-  if (engineMessages.length > MAX_CONTEXT_MESSAGES) {
-    return [engineMessages[0], ...engineMessages.slice(-(MAX_CONTEXT_MESSAGES - 1))];
-  }
-  return engineMessages;
-}
-
 export class CortexProvider extends BaseProvider {
   constructor() {
     super('cortex');
@@ -61,238 +42,74 @@ export class CortexProvider extends BaseProvider {
 
   async *stream(messages, options = {}) {
     const model = Catalog.getRealModel(options.model || Config.get('model'));
-    const engineMessages = buildEngineMessages(options.systemPrompt || '', messages);
+    const systemPrompt = options.systemPrompt || '';
 
-    const controller = new AbortController();
-    const connectTimer = setTimeout(() => controller.abort(), 600000);
-
-    let resp;
-    try {
-      resp = await fetch(`${this.host}/api/chat`, {
-        method: 'POST',
-        headers: { 'Content-Type': 'application/json' },
-        body: JSON.stringify({
-          model,
-          messages: engineMessages,
-          stream: true,
-          keep_alive: Config.get('keepAlive'),
-          options: {
-            num_ctx: Config.get('numCtx'),
-            num_predict: options.maxTokens ?? Config.get('maxTokens'),
-            temperature: options.temperature || Config.get('temperature'),
-            num_thread: Config.get('numThreads'),
-          },
-        }),
-        signal: controller.signal,
-      });
-    } catch (err) {
-      if (err.name === 'AbortError') {
-        throw new Error('[ERR-0001] Could not reach the engine (no response within 600s). Make sure the local engine is running.');
-      }
-      throw new Error('[ERR-0001] Could not reach the engine (no response). Make sure the local engine is running.');
-    } finally {
-      clearTimeout(connectTimer);
+    const engineMessages = [];
+    if (systemPrompt) {
+      engineMessages.push({ role: 'system', content: systemPrompt });
+    }
+    for (const msg of messages) {
+      engineMessages.push({ role: msg.role, content: msg.content });
     }
 
+    const resp = await fetch(`${this.host}/api/chat`, {
+      method: 'POST',
+      headers: { 'Content-Type': 'application/json' },
+      body: JSON.stringify({
+        model,
+        messages: engineMessages,
+        stream: true,
+        keep_alive: Config.get('keepAlive'),
+        options: {
+          num_ctx: Config.get('numCtx'),
+          temperature: options.temperature || Config.get('temperature'),
+        },
+      }),
+    });
+
     if (!resp.ok) {
-      throw new Error(await describeEngineError(resp, model));
+      throw new Error('Failed to connect to backend engine');
     }
 
     const reader = resp.body.getReader();
     const decoder = new TextDecoder();
     let buffer = '';
-    let idle = null;
-    const settle = () => {
-      if (idle) clearTimeout(idle);
-      idle = setTimeout(() => controller.abort(), 420000);
-    };
-    settle();
-    try {
-      while (true) {
-        let value, done;
-        try {
-          ({ done, value } = await reader.read());
-        } catch (err) {
-          if (err.name === 'AbortError') {
-            throw new Error('[ERR-0001] Lost the connection to the engine (no data for 420s).');
-          }
-          throw err;
-        }
-        if (done) break;
-        settle();
-        buffer += decoder.decode(value, { stream: true });
-        const lines = buffer.split('\n');
-        buffer = lines.pop() || '';
-
-        for (const line of lines) {
-          if (!line.trim()) continue;
-          try {
-            const json = JSON.parse(line);
-            if (json.message?.content) {
-              yield json.message.content;
-            }
-            if (json.done) return;
-          } catch {}
-        }
-      }
-    } finally {
-      if (idle) clearTimeout(idle);
-    }
-  }
-
-  // Non-streaming completion (used by the tool-planning step). Reads the
-  // streamed reply incrementally: on a slow engine a full answer can take
-  // minutes, so only a connection with no response for 180s (or an idle
-  // stream for 120s) counts as a failure - not a slowly progressing one.
-  async complete(messages, options = {}) {
-    const model = Catalog.getRealModel(options.model || Config.get('model'));
-    const engineMessages = buildEngineMessages(options.systemPrompt || '', messages);
-
-    const controller = new AbortController();
-    const connectTimer = setTimeout(() => controller.abort(), 600000);
-
-    if (options.signal) {
-      if (options.signal.aborted) controller.abort();
-      else options.signal.addEventListener('abort', () => controller.abort(), { once: true });
-    }
 
-    let resp;
-    try {
-      resp = await fetch(`${this.host}/api/chat`, {
-        method: 'POST',
-        headers: { 'Content-Type': 'application/json' },
-        body: JSON.stringify({
-          model,
-          messages: engineMessages,
-          stream: true,
-          keep_alive: Config.get('keepAlive'),
-          options: {
-            num_ctx: Config.get('numCtx'),
-            num_predict: options.maxTokens ?? Config.get('maxTokens'),
-            temperature: options.temperature || Config.get('temperature'),
-            num_thread: Config.get('numThreads'),
-          },
-        }),
-        signal: controller.signal,
-      });
-    } catch (err) {
-      if (err.name === 'AbortError') {
-        throw new Error('[ERR-0001] Could not complete the request (no response within 600s).');
-      }
-      throw new Error('[ERR-0001] Could not reach the engine (no response). Make sure the local engine is running.');
-    } finally {
-      clearTimeout(connectTimer);
-    }
+    while (true) {
+      const { done, value } = await reader.read();
+      if (done) break;
 
-    if (!resp.ok) {
-      throw new Error(await describeEngineError(resp, model));
-    }
+      buffer += decoder.decode(value, { stream: true });
+      const lines = buffer.split('\n');
+      buffer = lines.pop() || '';
 
-    const reader = resp.body.getReader();
-    const decoder = new TextDecoder();
-    let buffer = '';
-    let idle = null;
-    const settle = () => {
-      if (idle) clearTimeout(idle);
-      idle = setTimeout(() => controller.abort(), 120000);
-    };
-    settle();
-    let output = '';
-    try {
-      while (true) {
-        let value, done;
+      for (const line of lines) {
+        if (!line.trim()) continue;
         try {
-          ({ done, value } = await reader.read());
-        } catch (err) {
-          if (err.name === 'AbortError') {
-            throw new Error('[ERR-0001] Lost the connection to the engine (no data for 420s).');
+          const json = JSON.parse(line);
+          if (json.message?.content) {
+            yield json.message.content;
           }
-          throw err;
-        }
-        if (done) break;
-        settle();
-        buffer += decoder.decode(value, { stream: true });
-        const lines = buffer.split('\n');
-        buffer = lines.pop() || '';
-
-        for (const line of lines) {
-          if (!line.trim()) continue;
-          try {
-            const json = JSON.parse(line);
-            if (json.message?.content) {
-              output += json.message.content;
-            }
-            if (json.done) {
-              buffer = '';
-              break;
-            }
-          } catch {}
-        }
+          if (json.done) return;
+        } catch {}
       }
-    } finally {
-      if (idle) clearTimeout(idle);
-      controller.abort();
     }
-    return output;
   }
 
   async warmup() {
-    // Fire-and-forget: ask the engine to load the model eagerly, then abort as
-    // soon as generation starts. The engine runs with a single slot, so an
-    // in-flight non-streaming 'hi' would block the first real request.
+    // Fire-and-forget: send a tiny request so ollama loads the model into
+    // memory eagerly, so the first real query doesn't pay the cold-load cost.
     const model = Catalog.getRealModel(Config.get('model'));
-    const controller = new AbortController();
     fetch(`${this.host}/api/chat`, {
       method: 'POST',
       headers: { 'Content-Type': 'application/json' },
       body: JSON.stringify({
         model,
         messages: [{ role: 'user', content: 'hi' }],
-        stream: true,
+        stream: false,
         keep_alive: Config.get('keepAlive'),
         options: { num_ctx: Config.get('numCtx') },
       }),
-      signal: controller.signal,
-    })
-      .then(async (resp) => {
-        if (!resp.ok || !resp.body) return;
-        const reader = resp.body.getReader();
-        const decoder = new TextDecoder();
-        let buffer = '';
-        for (let i = 0; i < 100; i++) {
-          const { done, value } = await reader.read();
-          if (done) break;
-          buffer += decoder.decode(value, { stream: true });
-          const lines = buffer.split('\n');
-          buffer = lines.pop() || '';
-          for (const line of lines) {
-            if (!line.trim()) continue;
-            try {
-              if (JSON.parse(line).message?.content) {
-                controller.abort();
-                return;
-              }
-            } catch {}
-          }
-        }
-      })
-      .catch(() => {});
-  }
-}
-
-// Build a user-actionable message when the engine rejects a request (e.g. a
-// cloud model name posted to a local engine -> 404 "model not found"). The
-// engine host is deliberately kept out of the message; errors carry codes.
-async function describeEngineError(resp, model) {
-  if (/not found|does not exist|model.*missing/i.test(String(resp.statusText))) {
-    return `[ERR-0002] Model "${model}" is not installed on the engine. If you picked a cloud model, use /model VRTL-6.pro or /provider openai.`;
-  }
-  let body = '';
-  try {
-    body = (await resp.text()).slice(0, 300);
-  } catch {}
-  if (/not found|does not exist|model.*missing/i.test(body)) {
-    return `[ERR-0002] Model "${model}" is not installed on the engine. If you picked a cloud model, use /model VRTL-6.pro or /provider openai.`;
+    }).catch(() => {});
   }
-  return `[ERR-0003] The engine returned HTTP ${resp.status}. Check the engine status or run /model VRTL-2.fast.`;
 }

package/src/providers/gemini.js

@@ -39,7 +39,7 @@ export class GeminiProvider extends BaseProvider {
 
   async *stream(messages, options = {}) {
     const key = Config.get('geminiApiKey');
-    if (!key) throw new Error('[ERR-0004] Cloud API key not configured. Add it to the config and try again.');
+    if (!key) throw new Error('Gemini API key not configured');
 
     const model = this._resolveModel(options);
     const systemPrompt = options.systemPrompt || '';
@@ -52,77 +52,46 @@ export class GeminiProvider extends BaseProvider {
       contents.unshift({ role: 'user', parts: [{ text: `System: ${systemPrompt}` }] });
     }
 
-    const controller = new AbortController();
-    const connectTimer = setTimeout(() => controller.abort(), 240000);
-
-    let resp;
-    try {
-      resp = await fetch(
-        `${GEMINI_URL}/${model}:streamGenerateContent?key=${key}&alt=sse`,
-        {
-          method: 'POST',
-          headers: { 'Content-Type': 'application/json' },
-          body: JSON.stringify({
-            contents,
-            generationConfig: {
-              temperature: options.temperature || Config.get('temperature'),
-              maxOutputTokens: options.maxTokens || Config.get('maxTokens'),
-            },
-          }),
-          signal: controller.signal,
-        }
-      );
-    } catch (err) {
-      if (err.name === 'AbortError') {
-        throw new Error('[ERR-0001] Could not reach the API (no response within 60s). Check your network connection.');
+    const resp = await fetch(
+      `${GEMINI_URL}/${model}:streamGenerateContent?key=${key}&alt=sse`,
+      {
+        method: 'POST',
+        headers: { 'Content-Type': 'application/json' },
+        body: JSON.stringify({
+          contents,
+          generationConfig: {
+            temperature: options.temperature || Config.get('temperature'),
+            maxOutputTokens: options.maxTokens || Config.get('maxTokens'),
+          },
+        }),
       }
-      throw new Error('[ERR-0001] Could not reach the API (no response). Check your network connection.');
-    } finally {
-      clearTimeout(connectTimer);
-    }
+    );
 
     if (!resp.ok) {
-      throw new Error('[ERR-0003] The cloud provider returned HTTP ' + resp.status + '. Check your API key in config.');
+      throw new Error('Failed to connect to cloud engine');
     }
 
     const reader = resp.body.getReader();
     const decoder = new TextDecoder();
     let buffer = '';
-    let idle = null;
-    const settle = () => {
-      if (idle) clearTimeout(idle);
-      idle = setTimeout(() => controller.abort(), 240000);
-    };
-    settle();
-    try {
-      while (true) {
-        let value, done;
-        try {
-          ({ done, value } = await reader.read());
-        } catch (err) {
-          if (err.name === 'AbortError') {
-            throw new Error('[ERR-0001] Lost the connection to the cloud API (no data for 60s).');
-          }
-          throw err;
-        }
-        if (done) break;
-        settle();
-        buffer += decoder.decode(value, { stream: true });
-        const lines = buffer.split('\n');
-        buffer = lines.pop() || '';
 
-        for (const line of lines) {
-          if (!line.startsWith('data: ')) continue;
-          const data = line.slice(6);
-          try {
-            const json = JSON.parse(data);
-            const text = json.candidates?.[0]?.content?.parts?.[0]?.text;
-            if (text) yield text;
-          } catch {}
-        }
+    while (true) {
+      const { done, value } = await reader.read();
+      if (done) break;
+
+      buffer += decoder.decode(value, { stream: true });
+      const lines = buffer.split('\n');
+      buffer = lines.pop() || '';
+
+      for (const line of lines) {
+        if (!line.startsWith('data: ')) continue;
+        const data = line.slice(6);
+        try {
+          const json = JSON.parse(data);
+          const text = json.candidates?.[0]?.content?.parts?.[0]?.text;
+          if (text) yield text;
+        } catch {}
       }
-    } finally {
-      if (idle) clearTimeout(idle);
     }
   }
 }

package/src/providers/openai.js

@@ -57,7 +57,7 @@ export class OpenAIProvider extends BaseProvider {
 
   async *stream(messages, options = {}) {
     const key = Config.get('openaiApiKey');
-    if (!key) throw new Error('[ERR-0004] Cloud API key not configured. Add it to the config and try again.');
+    if (!key) throw new Error('Cloud API key not configured');
 
     const model = this._resolveModel(options);
     const systemPrompt = options.systemPrompt || '';
@@ -70,78 +70,47 @@ export class OpenAIProvider extends BaseProvider {
       apiMessages.push({ role: msg.role, content: msg.content });
     }
 
-    const controller = new AbortController();
-    const connectTimer = setTimeout(() => controller.abort(), 240000);
-
-    let resp;
-    try {
-      resp = await fetch('https://api.openai.com/v1/chat/completions', {
-        method: 'POST',
-        headers: {
-          'Content-Type': 'application/json',
-          Authorization: `Bearer ${key}`,
-        },
-        body: JSON.stringify({
-          model,
-          messages: apiMessages,
-          stream: true,
-          temperature: options.temperature || Config.get('temperature'),
-          max_tokens: options.maxTokens || Config.get('maxTokens'),
-        }),
-        signal: controller.signal,
-      });
-    } catch (err) {
-      if (err.name === 'AbortError') {
-        throw new Error('[ERR-0001] Could not reach the API (no response within 60s). Check your network connection.');
-      }
-      throw new Error('[ERR-0001] Could not reach the API (no response). Check your network connection.');
-    } finally {
-      clearTimeout(connectTimer);
-    }
+    const resp = await fetch('https://api.openai.com/v1/chat/completions', {
+      method: 'POST',
+      headers: {
+        'Content-Type': 'application/json',
+        Authorization: `Bearer ${key}`,
+      },
+      body: JSON.stringify({
+        model,
+        messages: apiMessages,
+        stream: true,
+        temperature: options.temperature || Config.get('temperature'),
+        max_tokens: options.maxTokens || Config.get('maxTokens'),
+      }),
+    });
 
     if (!resp.ok) {
-      throw new Error('[ERR-0003] The cloud provider returned HTTP ' + resp.status + '. Check your API key in config.');
+      throw new Error('Failed to connect to cloud engine');
     }
 
     const reader = resp.body.getReader();
     const decoder = new TextDecoder();
     let buffer = '';
-    let idle = null;
-    const settle = () => {
-      if (idle) clearTimeout(idle);
-      idle = setTimeout(() => controller.abort(), 240000);
-    };
-    settle();
-    try {
-      while (true) {
-        let value, done;
-        try {
-          ({ done, value } = await reader.read());
-        } catch (err) {
-          if (err.name === 'AbortError') {
-            throw new Error('[ERR-0001] Lost the connection to the cloud API (no data for 60s).');
-          }
-          throw err;
-        }
-        if (done) break;
-        settle();
-        buffer += decoder.decode(value, { stream: true });
-        const lines = buffer.split('\n');
-        buffer = lines.pop() || '';
 
-        for (const line of lines) {
-          if (!line.startsWith('data: ')) continue;
-          const data = line.slice(6);
-          if (data === '[DONE]') return;
-          try {
-            const json = JSON.parse(data);
-            const content = json.choices?.[0]?.delta?.content;
-            if (content) yield content;
-          } catch {}
-        }
+    while (true) {
+      const { done, value } = await reader.read();
+      if (done) break;
+
+      buffer += decoder.decode(value, { stream: true });
+      const lines = buffer.split('\n');
+      buffer = lines.pop() || '';
+
+      for (const line of lines) {
+        if (!line.startsWith('data: ')) continue;
+        const data = line.slice(6);
+        if (data === '[DONE]') return;
+        try {
+          const json = JSON.parse(data);
+          const content = json.choices?.[0]?.delta?.content;
+          if (content) yield content;
+        } catch {}
       }
-    } finally {
-      if (idle) clearTimeout(idle);
     }
   }
 }