COMMUNITY · PYTHON

Hermes Agent plugin (community)

hermes-dakera gives Hermes Agent memory that survives the session: every turn is retained, and relevant memories are recalled before the next one, on a Dakera server you run yourself.

Community-maintained, not an official Dakera package. Built by @Xyzjesus. We have not tested or reviewed it for support, and the repository did not declare a licence when we reviewed it in October 2026. Check the repository before depending on it. · GitHub →

What you get from the pairing

PERSISTENT ACROSS SESSIONS

Turns are stored as user and assistant pairs under an agent_id, so what Hermes learned yesterday is recallable today.

HYBRID RECALL

Recall routes to vector, bm25 or hybrid search (default auto) with a configurable top-K. Dakera fuses and reranks on the server.

MULTILINGUAL

A v0.12.0 Dakera server embeds 100+ languages in one space, so a question in one language can find a memory written in another. Details

ON YOUR OWN SERVER

Memory text goes to the Dakera container you run, with no per-call fee. If it is unreachable, the plugin fails open and Hermes keeps working without memory.

Hermes Agent is built by Nous Research, is MIT-licensed and had about 250k GitHub stars in October 2026 (source).

How it fits together

As described in the plugin README, two hooks do the work: a per-turn retain after each exchange, and a pre-turn recall that is prefetched on a background thread.

After each turn: retainwrite

Hermes agentfinishes a turn

Plugin hook: retainuser and assistant pair, noise-gated

Dakera serverembeds and stores

Memoriesscoped by agent_id

Before each turn: recallread

Hermes agentnext turn begins

Plugin hook: recallprefetched on a background thread

Dakera serverhybrid recall and rerank

Contextrecalled memories injected

Your infrastructure

Hook behaviour as documented in the hermes-dakera README. A session-end summary of the last six messages is also stored.

Tools

The agent also gets four tools it can call itself, as listed in the README.

Tool	What it does
`dakera_recall`	Semantic retrieval that returns memories with their scores
`dakera_store`	Persists information with an importance from 0 to 1 and a memory type
`dakera_search`	Browses memories by BM25 keyword matching
`dakera_forget`	Removes memories by identifier

Install in three steps

As documented in the project README:

Run the Dakera server

docker run -d -p 3000:3000 -e DAKERA_ROOT_API_KEY=YOUR-TOKEN ghcr.io/dakera-ai/dakera:latest

For a pinned version and a persistent data volume, see the Quickstart.

Install the plugin and write its config

hermes plugins install Xyzjesus/hermes-dakera
mkdir -p ~/.hermes/dakera
printf '{"api_url":"http://YOUR-DAKERA-HOST:3000","api_key":"YOUR-TOKEN","agent_id":"hermes","top_k":5,"recall_routing":"auto"}
'   > ~/.hermes/dakera/config.json && chmod 600 ~/.hermes/dakera/config.json

Activate and verify

Set the provider in ~/.hermes/config.yaml, then check its status:

memory: { provider: dakera }

hermes memory status

Compatibility

Item	Detail
Plugin version	0.1.0, first published 27 September 2026
Dakera Python SDK	Declares `dakera>=0.12.11,<0.13`, the 0.12 series. Dakera's own Python SDK for server v0.12.0 is the 0.13 line (currently 0.13.2), so check the repository for an update before pairing it with a v0.12.0 server.
Hermes	Hermes Agent by Nous Research (memory provider plugin)
Maintainer	@Xyzjesus, independent of Dakera

Links

Install in minutes

Add persistent memory to your framework.

One Docker command starts the server. One pip or npm install gives your agents memory that survives the session.

Read the quickstart → All integrations

Prefer managed hosting? Join the Dakera Cloud waitlist →