feat: Semantische Suche über Chat-History (memory_search)
- Lokales Embedding-Model (all-MiniLM-L6-v2, ONNX, CPU) - EmbeddingService: WordPiece-Tokenizer + Mean Pooling (384 Dim) - VectorSearchService: SQLite-basierte Vektor-Speicherung + Cosine Similarity - EmbeddingBackgroundService: Backfill bestehender Messages + Live-Queue - Neues Tool memory_search: Semantische Suche über alle Sessions (inkl. archivierte) - ~9000 Messages embedded, Suche über komplette Chat-History
This commit is contained in:
Meleki committed
1 parent
baf53b7dac
commit
add111ffe1
9 files changed
+727
-15
No files matched your search
@@ -0,0 +1,214 @@
|
||||
using System;
|
||||
using System.Collections.Generic;
|
||||
using System.Linq;
|
||||
using System.Threading;
|
||||
using System.Threading.Tasks;
|
||||
using Dapper;
|
||||
using Microsoft.Data.Sqlite;
|
||||
|
||||
namespace Gott.Services;
|
||||
|
||||
/// <summary>
|
||||
/// Verwaltet Message-Embeddings in SQLite und bietet semantische Suche.
|
||||
/// </summary>
|
||||
public class VectorSearchService
|
||||
{
|
||||
readonly string _connectionString;
|
||||
readonly EmbeddingService _embedding;
|
||||
readonly SemaphoreSlim _embeddingLock = new(1, 1);
|
||||
|
||||
public VectorSearchService(string dbPath, EmbeddingService embedding)
|
||||
{
|
||||
_connectionString = $"Data Source={dbPath}";
|
||||
_embedding = embedding;
|
||||
InitializeSchema();
|
||||
}
|
||||
|
||||
SqliteConnection CreateConnection()
|
||||
{
|
||||
var conn = new SqliteConnection(_connectionString);
|
||||
conn.Open();
|
||||
return conn;
|
||||
}
|
||||
|
||||
void InitializeSchema()
|
||||
{
|
||||
using var conn = CreateConnection();
|
||||
conn.Execute(@"
|
||||
CREATE TABLE IF NOT EXISTS message_embeddings (
|
||||
id INTEGER PRIMARY KEY AUTOINCREMENT,
|
||||
message_id INTEGER NOT NULL UNIQUE,
|
||||
chain_id TEXT NOT NULL,
|
||||
session_id TEXT NOT NULL,
|
||||
agent_id TEXT,
|
||||
role TEXT NOT NULL,
|
||||
text_preview TEXT,
|
||||
vector BLOB NOT NULL,
|
||||
created_at TEXT NOT NULL
|
||||
);
|
||||
CREATE INDEX IF NOT EXISTS idx_emb_session ON message_embeddings(session_id);
|
||||
CREATE INDEX IF NOT EXISTS idx_emb_agent ON message_embeddings(agent_id);
|
||||
CREATE INDEX IF NOT EXISTS idx_emb_message ON message_embeddings(message_id);
|
||||
");
|
||||
Console.WriteLine("[VectorSearch] Schema initialized.");
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Generiert und speichert ein Embedding für eine Message.
|
||||
/// Thread-safe durch SemaphoreSlim (ONNX InferenceSession ist nicht thread-safe).
|
||||
/// </summary>
|
||||
public async Task EmbedMessageAsync(long messageId, string chainId, string sessionId,
|
||||
string agentId, string role, string text)
|
||||
{
|
||||
if (!_embedding.IsAvailable || string.IsNullOrWhiteSpace(text)) return;
|
||||
if (text.Length < 10) return; // Zu kurze Texte sind nicht sinnvoll
|
||||
|
||||
// Prüfe ob schon embedded
|
||||
using var conn = CreateConnection();
|
||||
var exists = await conn.ExecuteScalarAsync<int>(
|
||||
"SELECT COUNT(*) FROM message_embeddings WHERE message_id = @messageId",
|
||||
new { messageId });
|
||||
if (exists > 0) return;
|
||||
|
||||
// Embedding generieren (thread-safe)
|
||||
float[] vector;
|
||||
await _embeddingLock.WaitAsync();
|
||||
try
|
||||
{
|
||||
vector = _embedding.GenerateEmbedding(text);
|
||||
}
|
||||
finally
|
||||
{
|
||||
_embeddingLock.Release();
|
||||
}
|
||||
|
||||
if (vector.Length == 0) return;
|
||||
|
||||
var preview = text.Length > 200 ? text.Substring(0, 200) : text;
|
||||
|
||||
await conn.ExecuteAsync(@"
|
||||
INSERT OR IGNORE INTO message_embeddings
|
||||
(message_id, chain_id, session_id, agent_id, role, text_preview, vector, created_at)
|
||||
VALUES
|
||||
(@messageId, @chainId, @sessionId, @agentId, @role, @preview, @vector, @createdAt)",
|
||||
new
|
||||
{
|
||||
messageId,
|
||||
chainId,
|
||||
sessionId,
|
||||
agentId = agentId ?? "",
|
||||
role,
|
||||
preview,
|
||||
vector = EmbeddingService.ToBytes(vector),
|
||||
createdAt = DateTime.UtcNow.ToString("o")
|
||||
});
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Semantische Suche: Findet die N ähnlichsten Messages zu einer Query.
|
||||
/// Durchsucht alle Sessions eines Agents (oder alle wenn agentId null).
|
||||
/// </summary>
|
||||
public async Task<List<SearchResult>> SearchAsync(string query, string agentId = null,
|
||||
int topN = 10, float minScore = 0.3f)
|
||||
{
|
||||
if (!_embedding.IsAvailable) return new List<SearchResult>();
|
||||
|
||||
// Query embedden
|
||||
float[] queryVec;
|
||||
await _embeddingLock.WaitAsync();
|
||||
try
|
||||
{
|
||||
queryVec = _embedding.GenerateEmbedding(query);
|
||||
}
|
||||
finally
|
||||
{
|
||||
_embeddingLock.Release();
|
||||
}
|
||||
|
||||
if (queryVec.Length == 0) return new List<SearchResult>();
|
||||
|
||||
// Alle Embeddings laden und Cosine Similarity berechnen
|
||||
using var conn = CreateConnection();
|
||||
string sql;
|
||||
object param;
|
||||
|
||||
if (!string.IsNullOrEmpty(agentId))
|
||||
{
|
||||
sql = "SELECT message_id, chain_id, session_id, agent_id, role, text_preview, vector, created_at FROM message_embeddings WHERE agent_id = @agentId";
|
||||
param = new { agentId };
|
||||
}
|
||||
else
|
||||
{
|
||||
sql = "SELECT message_id, chain_id, session_id, agent_id, role, text_preview, vector, created_at FROM message_embeddings";
|
||||
param = new { };
|
||||
}
|
||||
|
||||
var rows = await conn.QueryAsync<EmbeddingRow>(sql, param);
|
||||
var results = new List<SearchResult>();
|
||||
|
||||
foreach (var row in rows)
|
||||
{
|
||||
var vec = EmbeddingService.FromBytes(row.Vector);
|
||||
var score = EmbeddingService.CosineSimilarity(queryVec, vec);
|
||||
if (score >= minScore)
|
||||
{
|
||||
results.Add(new SearchResult
|
||||
{
|
||||
MessageId = row.MessageId,
|
||||
ChainId = row.ChainId,
|
||||
SessionId = row.SessionId,
|
||||
AgentId = row.AgentId,
|
||||
Role = row.Role,
|
||||
TextPreview = row.TextPreview,
|
||||
Score = score,
|
||||
CreatedAt = row.CreatedAt
|
||||
});
|
||||
}
|
||||
}
|
||||
|
||||
return results.OrderByDescending(r => r.Score).Take(topN).ToList();
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Gibt die Anzahl der gespeicherten Embeddings zurück.
|
||||
/// </summary>
|
||||
public int GetEmbeddingCount()
|
||||
{
|
||||
using var conn = CreateConnection();
|
||||
return conn.ExecuteScalar<int>("SELECT COUNT(*) FROM message_embeddings");
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Gibt die IDs aller bereits eingebetteten Messages zurück.
|
||||
/// </summary>
|
||||
public async Task<HashSet<long>> GetEmbeddedMessageIdsAsync()
|
||||
{
|
||||
using var conn = CreateConnection();
|
||||
var ids = await conn.QueryAsync<long>("SELECT message_id FROM message_embeddings");
|
||||
return new HashSet<long>(ids);
|
||||
}
|
||||
|
||||
public class EmbeddingRow
|
||||
{
|
||||
public long MessageId { get; set; }
|
||||
public string ChainId { get; set; }
|
||||
public string SessionId { get; set; }
|
||||
public string AgentId { get; set; }
|
||||
public string Role { get; set; }
|
||||
public string TextPreview { get; set; }
|
||||
public byte[] Vector { get; set; }
|
||||
public string CreatedAt { get; set; }
|
||||
}
|
||||
|
||||
public class SearchResult
|
||||
{
|
||||
public long MessageId { get; set; }
|
||||
public string ChainId { get; set; }
|
||||
public string SessionId { get; set; }
|
||||
public string AgentId { get; set; }
|
||||
public string Role { get; set; }
|
||||
public string TextPreview { get; set; }
|
||||
public float Score { get; set; }
|
||||
public string CreatedAt { get; set; }
|
||||
}
|
||||
}
|
||||
Reference in new issue
Block a user