MemoryRouter
Hermes Agent + MemoryRouter

New session.
Same memory.

Hermes Agent is the open-source AI agent from Nous Research. Give it persistent conversational memory, so what you told it yesterday can be there tomorrow.

Save my memories Install the plugin

Native plugin. Open source. Tested on Hermes 0.19.0.

01 / Keep the conversation

You already
explained this.

When your agent forgets what you told it yesterday, today starts with a recap. The decision. The constraints. The detail that made the answer useful.

Hermes already keeps persistent memory files. MemoryRouter adds the conversation around them: clean exchanges captured automatically, with relevant context recalled before the next turn. Less manual curation. Less starting over.

02 / Three commands

Give Hermes a longer memory.

Already running Hermes? Get your MemoryRouter key with the button above and export MEMORYROUTER_API_KEY in the shell where Hermes runs. Then use the same Python environment as Hermes:

pip install hermes-memoryrouter
hermes-memoryrouter install
hermes config set memory.provider memoryrouter

Hermes 0.19.0 needs Python 3.11 through 3.13. The installer puts the provider in $HERMES_HOME/plugins/memoryrouter/, or ~/.hermes/plugins/memoryrouter/ by default. Re-run it after package upgrades.

Full setup and key-file option GitHub PyPI 0.1.1
03 / Before and after every turn

Conversation in.
Context back.

  1. 01 / Recall

    Start with what matters.

    Before a turn, the provider retrieves relevant memories through /v1/memory/prepare and adds bounded background context.

  2. 02 / Respond

    Keep your model.

    Hermes uses your existing model provider and inference credentials. MemoryRouter supplies memory, not a new model connection.

  3. 03 / Capture

    Keep the clean exchange.

    After the turn, your words and the final reply go to /v1/memory/ingest. Tool and thinking noise are filtered out.

Memory is stored in the account behind your key. Only enable capture for conversations you want stored there. Recall can add up to its configured timeout; failed memory requests do not crash Hermes.

04 / Tested before it shipped

Not just an
import test.

32 offline tests passed. Then two real Hermes conversations ran in separate processes: one taught a unique token, the next recalled it exactly. No token in the recall question. No shared conversation history.

After publishing, we installed Hermes and hermes-memoryrouter from PyPI in another fresh environment and repeated the proof with a new token. Exact recall again.

Verified September 8, 2026. Hermes 0.19.0, plugin 0.1.1, Python 3.12, GPT-4.1 mini. This proves the tested setup, not every model or deployment.

Read the release story
05 / The practical details

A few good questions.

Your context. Still yours.

Pick up where
you left off.

Connect your memory. Start a conversation. Give the next session something to remember.

Save my memories