CODEX_ADK/BACKEND/Codex.Dal/Services/OllamaService.cs

using Codex.Dal.Entities;
using Codex.Dal.Enums;
using System.Net.Http.Json;
using System.Text;
using System.Text.Json.Serialization;

namespace Codex.Dal.Services;

/// <summary>
/// Implementation of Ollama service for LLM interactions
/// </summary>
public class OllamaService : IOllamaService
{
    private readonly IHttpClientFactory _httpClientFactory;

    public OllamaService(IHttpClientFactory httpClientFactory)
    {
        _httpClientFactory = httpClientFactory;
    }

    public async Task<OllamaResponse> GenerateAsync(
        string endpoint,
        string model,
        string systemPrompt,
        List<ConversationMessage> contextMessages,
        string userMessage,
        double temperature,
        int maxTokens,
        CancellationToken cancellationToken = default)
    {
        var httpClient = _httpClientFactory.CreateClient();
        httpClient.Timeout = TimeSpan.FromMinutes(5); // Allow for longer generation times

        // Build prompt with system instruction and conversation context
        var promptBuilder = new StringBuilder();

        // Add system prompt
        if (!string.IsNullOrWhiteSpace(systemPrompt))
        {
            promptBuilder.AppendLine($"System: {systemPrompt}");
            promptBuilder.AppendLine();
        }

        // Add conversation context
        foreach (var msg in contextMessages)
        {
            var role = msg.Role switch
            {
                MessageRole.User => "User",
                MessageRole.Assistant => "Assistant",
                MessageRole.System => "System",
                MessageRole.Tool => "Tool",
                _ => "Unknown"
            };

            promptBuilder.AppendLine($"{role}: {msg.Content}");
        }

        // Add current user message
        promptBuilder.AppendLine($"User: {userMessage}");
        promptBuilder.Append("Assistant:");

        // Build request payload
        var payload = new OllamaGenerateRequest
        {
            Model = model,
            Prompt = promptBuilder.ToString(),
            Temperature = temperature,
            Options = new OllamaOptions
            {
                NumPredict = maxTokens,
                Temperature = temperature
            },
            Stream = false
        };

        try
        {
            var response = await httpClient.PostAsJsonAsync(
                $"{endpoint.TrimEnd('/')}/api/generate",
                payload,
                cancellationToken
            );

            response.EnsureSuccessStatusCode();

            var result = await response.Content.ReadFromJsonAsync<OllamaGenerateResponse>(cancellationToken);

            if (result == null)
            {
                throw new InvalidOperationException("Received null response from Ollama API");
            }

            return new OllamaResponse
            {
                Content = result.Response?.Trim() ?? string.Empty,
                InputTokens = result.PromptEvalCount,
                OutputTokens = result.EvalCount
            };
        }
        catch (HttpRequestException ex)
        {
            throw new InvalidOperationException(
                $"Failed to connect to Ollama at {endpoint}. Ensure Ollama is running and the endpoint is correct.",
                ex
            );
        }
        catch (TaskCanceledException ex)
        {
            throw new InvalidOperationException(
                $"Request to Ollama timed out. The model may be taking too long to respond.",
                ex
            );
        }
    }

    /// <summary>
    /// Request payload for Ollama /api/generate endpoint
    /// </summary>
    private record OllamaGenerateRequest
    {
        [JsonPropertyName("model")]
        public string Model { get; init; } = string.Empty;

        [JsonPropertyName("prompt")]
        public string Prompt { get; init; } = string.Empty;

        [JsonPropertyName("temperature")]
        public double Temperature { get; init; }

        [JsonPropertyName("options")]
        public OllamaOptions? Options { get; init; }

        [JsonPropertyName("stream")]
        public bool Stream { get; init; }
    }

    /// <summary>
    /// Options for Ollama generation
    /// </summary>
    private record OllamaOptions
    {
        [JsonPropertyName("num_predict")]
        public int NumPredict { get; init; }

        [JsonPropertyName("temperature")]
        public double Temperature { get; init; }
    }

    /// <summary>
    /// Response from Ollama /api/generate endpoint
    /// </summary>
    private record OllamaGenerateResponse
    {
        [JsonPropertyName("response")]
        public string? Response { get; init; }

        [JsonPropertyName("model")]
        public string? Model { get; init; }

        [JsonPropertyName("created_at")]
        public string? CreatedAt { get; init; }

        [JsonPropertyName("done")]
        public bool Done { get; init; }

        [JsonPropertyName("total_duration")]
        public long? TotalDuration { get; init; }

        [JsonPropertyName("load_duration")]
        public long? LoadDuration { get; init; }

        [JsonPropertyName("prompt_eval_count")]
        public int? PromptEvalCount { get; init; }

        [JsonPropertyName("prompt_eval_duration")]
        public long? PromptEvalDuration { get; init; }

        [JsonPropertyName("eval_count")]
        public int? EvalCount { get; init; }

        [JsonPropertyName("eval_duration")]
        public long? EvalDuration { get; init; }
    }
}