Pydantic AI + MemorySync

Overview

Install pydantic-ai-memorysync and add MemorySyncCapability(user_id=...) to Agent(capabilities=[...]): recall injects through the framework’s instructions channel — never a conversation part, so message-history replay can never stack stale memory blocks — and turns persist in after_run, which fires only on success, so a failed run leaves no half-remembered turn. Identity resolves from your deps for multi-user servers. Five structured tools plus a typed search tool returning validated Pydantic models. The only memory integration built on pydantic-ai v2’s capability system.

Setup status and requirements

Supported version

pydantic-ai-memorysync 1.0.0 (PyPI)

Last setup review

2026-08-23

Permissions

A MemorySync API key with read and write scopes; the delete tool needs delete permission.

Limits

Requires pydantic-ai (or pydantic-ai-slim) >=2 <3 (Python 3.10+). Python only — TypeScript agents use the Vercel AI SDK or Mastra integrations.

Capabilities

• Built on the pydantic-ai v2 capability system (competitors pin <2)
• Recall rides the instructions channel — replay-safe by construction
• One recall per run, replayed across multi-request tool-calling runs
• Success-only persistence: failed runs leave no half-turns
• Identity from your deps: user_id_resolver or ctx.deps.user_id
• Typed search results: list[MemoryResult] Pydantic models
• Five never-raise agent tools with a read-only mode

Quick Start

from pydantic_ai import Agent
from pydantic_ai_memorysync import MemorySyncCapability

agent = Agent(
    "openai:gpt-5",
    instructions="You are a helpful assistant.",
    capabilities=[MemorySyncCapability(user_id="customer-7")],
)

result = await agent.run("What should I cook tonight?")

Use Cases

Explore More