Skip to content

← All guides

Archive memory guide

This integration guide is preserved for existing users. Memory is now a supporting primitive under Novyx change control, not the main product path.

How to Add Persistent Memory to Aider

Aider is one of the best terminal-based AI coding tools, but it has no built-in persistent memory. Every session starts fresh. With the Novyx SDK, you can wrap Aider to store and recall project context across sessions — so your AI pair programmer actually learns from your codebase over time.

The Problem

Aider excels at editing code from the terminal, but each session is isolated.

  • Aider starts every session with zero memory of previous interactions
  • You re-explain project architecture, conventions, and preferences every time
  • No built-in MCP support — you cannot plug in memory servers directly
  • No way to undo if Aider internalizes a wrong assumption about your code
  • Context from one project never transfers to another

The Solution: Novyx

Since Aider does not support MCP, you use the Novyx Python SDK directly. A simple wrapper script stores session summaries after each Aider run and loads relevant context before the next one. Your Aider sessions build up institutional knowledge over time.

Quick Start

Install the Novyx SDK:

bash
pip install novyx

Create a wrapper script that gives Aider persistent memory:

python
#!/usr/bin/env python3
"""aider-with-memory.py — Aider wrapper with Novyx persistent memory."""
import subprocess, sys
from novyx import NovyxClient

novyx = NovyxClient(api_key="YOUR_API_KEY")

# Recall relevant context before the session
project = sys.argv[1] if len(sys.argv) > 1 else "default"
memories = novyx.memory.search(
    query=f"project:{project} architecture conventions decisions",
    top_k=10,
)

# Build context string for Aider
context_lines = [m["observation"] for m in memories]
context = "\n".join(context_lines) if context_lines else "No prior context."

print(f"Loaded {len(context_lines)} memories for project: {project}")

# Run Aider with context injected via --message
subprocess.run([
    "aider",
    "--message", f"Context from previous sessions:\n{context}",
    *sys.argv[2:],
])

# After session, store a summary
summary = input("Session summary (or press Enter to skip): ").strip()
if summary:
    novyx.memory.store(
        observation=summary,
        tags=["aider", f"project:{project}", "session-summary"],
    )
    print("Session summary stored to Novyx.")

What You Get

Cross-Session Memory

Aider sessions build on each other. Yesterday's architecture decisions inform today's code changes automatically.

Semantic Search

The wrapper searches memories by meaning, not keywords. Ask about "database layer" and get results about "PostgreSQL schema decisions."

Project Isolation

Tag memories by project. Aider working on your API does not recall memories from your frontend project unless you want it to.

Recovery Support

Stored something wrong? Preview supported memory recovery and inspect the tamper-evident audit record.

How It Works

1

Install the SDK

pip install novyx — one dependency, no server to run.

2

Create the wrapper script

A thin Python script that calls novyx.memory.search() before Aider and novyx.memory.store() after.

3

Run Aider through the wrapper

python aider-with-memory.py myproject -- your usual aider flags here.

4

Memory accumulates automatically

Each session summary gets stored. Future sessions recall the most relevant context.

Frequently Asked Questions

Does Aider support MCP servers?

Not yet. Aider does not currently support the Model Context Protocol. You can integrate Novyx directly using the Python SDK in a wrapper script that runs alongside Aider.

How does Novyx work with Aider?

You write a thin Python wrapper that calls novyx.memory.store() after each Aider session and novyx.memory.search() before each session to load relevant context. The wrapper passes recalled memories into Aider as context.

Can I share memory between Aider and other tools?

Yes. Novyx is a standalone memory service. Memories stored during Aider sessions are available from Cursor, Claude Code, LangChain, or any other tool that connects to the same Novyx API key.

When Aider adds MCP support, you can switch to the Novyx MCP Server for a tighter integration. Until then, the SDK wrapper provides full memory capabilities.

Next protected action

Name the production action your agents cannot perform yet

Start with a 25-minute readiness call for one blocked workflow.

Book a readiness call

Name the production action your coding agents are not allowed to perform yet.

Email us at:

sales@novyxlabs.com

Include the agent tool, blocked production action, decision deadline, and the budget owner who should join. We will reply with fit and scheduling options for a 25-minute call.