YantrikDB Hermes Plugin
Memory provider for Hermes Agent: duplicates are merged, contradictions surface and every recalled memory carries a reason.
- 96 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The plugin stores conversation content in local memory and injects it into the agent requests to the model.
Install
- Works in
Manual install
pip install yantrikdb-hermes-pluginThen `yantrikdb-hermes install` and `hermes memory setup`, selecting yantrikdb.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The plugin stores conversation content in local memory and injects it into the agent requests to the model.
This is third-party code. Review the repository files before installing.
What it does
The plugin connects YantrikDB to Hermes Agent as a long-term memory provider. The engine runs in-process, so no separate server or token is needed. The agent stores facts in one session and recalls them in another, and every `yantrikdb_recall` result carries a `why_retrieved` reason. The memory maintains itself: it merges duplicates, exposes contradicting claims through `conflicts()`, notices knowledge gaps and turns recurring poorly answered queries into tasks. Before context compression the `on_pre_compress` hook injects the most salient memories so constraints survive long sessions. Rules in `yantrikdb-constitution.md` are injected first and outrank recalled memories.
Who it is for. Hermes Agent users with long sessions where constraints and preferences vanish after context compression.
Good fit when
- The agent has to remember your editor theme in the next session.
- An old claim is superseded by a new one and you need to see which one won.
- The agent runs around the clock on a mini PC or a Telegram gateway.
- A shared bot serves several people and facts of one must not reach another.
Not a fit when
- You need memory for Claude Code or Cursor, which have a separate yantrikdb-mcp server.
- A notes file without ranking and search is enough.
- The Python environment is older than 3.11.
Example request
Remember that I prefer a dark theme in VS Code, and in the next session ask which theme I like.Limitations
The plugin works only inside Hermes Agent and installs into the same Python environment as Hermes. Python 3.11–3.14 and YantrikDB engine versions from 0.12.1 up to 0.24 are supported. Memory separation between people is off by default because it changes where new memories are written, and it is enabled with `YANTRIKDB_OWNER_SCOPING=true`. An HTTP backend exists for high-availability clusters. The gap-to-task loop is on by default since version 0.10.0 and is disabled with `YANTRIKDB_AUTO_GAP_TASKS=false` and `YANTRIKDB_SURFACE_AGENDA=false`.
How to disable. Switch the provider off in the Hermes memory settings and remove the package with `pip uninstall yantrikdb-hermes-plugin`. Memory files live in the Hermes home directory and are removed separately.
FAQ
Do I need a separate server?
No. Since version 0.2.0 the engine runs inside the Hermes process and needs no server, token or GPU. The HTTP backend is for high-availability clusters only.
How do I see contradictions in memory?
Call `conflicts()`: it lists claims that disagree. Resolve one with `resolve_conflict()`.
What does the on_pre_compress hook do?
Before context compression it injects the most salient memories, so constraints set earlier survive a long session.
How do I set standing rules?
Write them into `$HERMES_HOME/yantrikdb-constitution.md`. The plugin injects them first and they outrank recalled memories.
Is there a version for Claude Code?
The same engine ships as the yantrikdb-mcp MCP server for Claude Code, Cursor and other clients and as a plain Python library.
Discussion
No questions or reviews yet.
Related
Pluginshermes-dreaming stages Hermes memory and skill updates for review and writes them to live state only after approval.
A pack of 19 plugins for Hermes Desktop: provider quotas, an iteration meter, a memory review queue, session management and RSS.
External memory for Hermes, Claude Code, Codex and Cursor: session archive, recall layers and relevant context injected back
Hermes Agent spends fewer tokens at start: the plugin drops unneeded toolsets and rolls back to the full set when unsure.
Agent memory in Rust: forgets the unimportant, merges similar items, flags contradictions, runs locally, over HTTP and via MCP.
Official Hermes Agent plugins: Home Assistant, Spotify, Blender, Snyk, Langfuse, sandboxes and importers for other agents
README in short
The repository is the canonical distribution: per Hermes maintainer guidance, new memory providers are not merged upstream, so the plugin installs through pip. It ships with benchmarks, demos and over 500 tests.