MemoryRouter
Aider memory

Aider memory: conventions files, then nothing after the session

Aider is built around explicit context: you add files, it edits them, the repository is the truth. That design is excellent for patches and poor for the reasons behind them.

Aider. Sources checked 2026-09-25.

How Aider handles memory today

Aider documents CONVENTIONS.md style files that you load with /read or the --read flag, so coding guidelines become read-only context for the session. The repo map gives the model a condensed view of the code it is working in.

There is no cross-session memory store. A new session starts from files and the repo map, so the discussion that shaped a change is gone even when the change is still in the tree.

What Aider forgets

Aider remembers code, not the argument about the code.

The fix with MemoryRouter

Aider speaks the OpenAI-compatible protocol, so MemoryRouter runs as a drop-in endpoint. Every request is retrieved and stored automatically, with no plugin to install.

  1. 1

    Point Aider at the MemoryRouter endpoint

    Shell environment

    export OPENAI_API_BASE=https://api.memoryrouter.ai/v1
    export OPENAI_API_KEY=mk_your_memory_key
  2. 2

    Start Aider as usual

    Run Aider with a MemoryRouter-supported model. Aider talks to the endpoint exactly like it talks to OpenAI, and memory happens inside the request path.

    Run Aider

    aider --model openai/gpt-5.5

Because this is a proxy, it works the same for any other OpenAI-compatible tool you already use.

Set up Aider with MemoryRouter

Exact commands and model notes on the Aider page. 14-day free trial.

Related

Aider memory FAQ

Sources

Sources checked 2026-09-25. Tool behavior changes; check the vendor docs before relying on a detail.

Start free