AI Agents & MCP: Give Your Agents Persistent Multimodal Memory

1import pixeltable as pxt
2from pixeltable.functions.huggingface import sentence_transformer
3
4pxt.create_dir('app', if_exists='ignore')
5
6# Persistent agent memory
7memory = pxt.create_table('app.agent_memory', {
8    'session_id': pxt.String,
9    'role': pxt.String,        # 'user', 'assistant', 'tool'
10    'content': pxt.String,
11    'timestamp': pxt.Timestamp,
12    'metadata': pxt.Json,
13}, if_exists='ignore')
14
15# Semantic search over memory; the index updates on every insert
16memory.add_embedding_index(
17    'content',
18    embedding=sentence_transformer.using(
19        model_id='sentence-transformers/all-MiniLM-L6-v2'
20    ),
21    if_exists='ignore',
22)
23
24# Agents can now search their own history
25sim = memory.content.similarity(string='what did we discuss about pricing?')
26relevant = (
27    memory.order_by(sim, asc=False)
28    .select(memory.content, memory.role, memory.session_id)
29    .limit(10)
30    .collect()
31)