dev: normalize Mistral, get tool calls working

Mistral has a lot of unnecessary quirks because the SDK coerces to
snake_case. We should see if the OpenAI client works with Mistral, and
it it does we'll use that instead so we don't have to maintain all these
exceptional cases.

src/backend/src/modules/puterai/MistralAIService.js

@@ -24,6 +24,7 @@ const { TypedValue } = require("../../services/drivers/meta/Runtime");
 const { nou } = require("../../util/langutil");
 
 const axios = require('axios');
+const OpenAIUtil = require("./lib/OpenAIUtil");
 const { TeePromise } = require('@heyputer/putility').libs.promise;
 
 
@@ -215,71 +216,51 @@ class MistralAIService extends BaseService {
              * AI Chat completion method.
              * See AIChatService for more details.
              */
-            async complete ({ messages, stream, model }) {
+            async complete ({ messages, stream, model, tools }) {
 
-                for ( let i = 0; i < messages.length; i++ ) {
-                    const message = messages[i];
-                    if ( ! message.role ) message.role = 'user';
+                messages = await OpenAIUtil.process_input_messages(messages);
+                for ( const message of messages ) {
+                    if ( message.tool_calls ) {
+                        message.toolCalls = message.tool_calls;
+                        delete message.tool_calls;
+                    }
+                    if ( message.tool_call_id ) {
+                        message.toolCallId = message.tool_call_id;
+                        delete message.tool_call_id;
+                    }
                 }
 
-                if ( stream ) {
-                    let usage_promise = new TeePromise();
+                console.log('MESSAGES TO MISTRAL', messages);
 
-                    const stream = new PassThrough();
-                    const retval = new TypedValue({
-                        $: 'stream',
-                        content_type: 'application/x-ndjson',
-                        chunked: true,
-                    }, stream);
-                    const completion = await this.client.chat.stream({
-                        model: model ?? this.get_default_model(),
-                        messages,
-                    });
-                    (async () => {
-                        for await ( let chunk of completion ) {
-                            // just because Mistral wants to be different
-                            chunk = chunk.data;
-
-                            if ( chunk.usage ) {
-                                usage_promise.resolve({
-                                    input_tokens: chunk.usage.promptTokens,
-                                    output_tokens: chunk.usage.completionTokens,
-                                });
-                                continue;
-                            }
-
-                            if ( chunk.choices.length < 1 ) continue;
-                            if ( chunk.choices[0].finish_reason ) {
-                                stream.end();
-                                break;
-                            }
-                            if ( nou(chunk.choices[0].delta.content) ) continue;
-                            const str = JSON.stringify({
-                                text: chunk.choices[0].delta.content
-                            });
-                            stream.write(str + '\n');
-                        }
-                        stream.end();
-                    })();
-
-                    return new TypedValue({ $: 'ai-chat-intermediate' }, {
-                        stream: true,
-                        response: retval,
-                        usage_promise: usage_promise,
-                    });
-                }
-
-                const completion = await this.client.chat.complete({
+                const completion = await this.client.chat[
+                    stream ? 'stream' : 'complete'
+                ]({
                     model: model ?? this.get_default_model(),
+                    ...(tools ? { tools } : {}),
                     messages,
                 });
-                // Expected case when mistralai/client-ts#23 is fixed
-                const ret = completion.choices[0];
-                ret.usage = {
-                    input_tokens: completion.usage.promptTokens,
-                    output_tokens: completion.usage.completionTokens,
-                };
-                return ret;
+            
+                return await OpenAIUtil.handle_completion_output({
+                    deviations: {
+                        index_usage_from_stream_chunk: chunk => {
+                            if ( ! chunk.usage ) return;
+
+                            const snake_usage = {};
+                            for ( const key in chunk.usage ) {
+                                const snakeKey = key.replace(/([A-Z])/g, "_$1").toLowerCase();
+                                snake_usage[snakeKey] = chunk.usage[key];
+                            }
+
+                            return snake_usage;
+                        },
+                        chunk_but_like_actually: chunk => chunk.data,
+                        index_tool_calls_from_stream_choice: choice => choice.delta.toolCalls,
+                    },
+                    completion, stream,
+                    usage_calculator: OpenAIUtil.create_usage_calculator({
+                        model_details: this.models_array_.find(m => m.id === model),
+                    }),
+                });
             }
         }
     }

src/backend/src/modules/puterai/lib/OpenAIUtil.js

@@ -84,7 +84,11 @@ module.exports = class OpenAIUtil {
         completion, usage_promise,
     }) => async ({ chatStream }) => {
         deviations = Object.assign({
+            // affected by: Groq
             index_usage_from_stream_chunk: chunk => chunk.usage,
+            // affected by: Mistral
+            chunk_but_like_actually: chunk => chunk,
+            index_tool_calls_from_stream_choice: choice => choice.delta.tool_calls,
         }, deviations);
 
         const message = chatStream.message();
@@ -94,7 +98,8 @@ module.exports = class OpenAIUtil {
         const tool_call_blocks = [];
 
         let last_usage = null;
-        for await ( const chunk of completion ) {
+        for await ( let chunk of completion ) {
+            chunk = deviations.chunk_but_like_actually(chunk);
             if ( process.env.DEBUG ) {
                 const delta = chunk?.choices?.[0]?.delta;
                 console.log(
@@ -119,12 +124,13 @@ module.exports = class OpenAIUtil {
                 continue;
             }
 
-            if (  ! nou(choice.delta.tool_calls) ) {
+            const tool_calls = deviations.index_tool_calls_from_stream_choice(choice);
+            if (  ! nou(tool_calls) ) {
                 if ( mode === 'text' ) {
                     mode = 'tool';
                     textblock.end();
                 }
-                for ( const tool_call of choice.delta.tool_calls ) {
+                for ( const tool_call of tool_calls ) {
                     if ( ! tool_call_blocks[tool_call.index] ) {
                         toolblock = message.contentBlock({
                             type: 'tool_use',