Hermes Agent

Add persistent long-term memory, Wiki retrieval, transcript capture, and built-in memory mirroring to Hermes Agent with Tvix.

Hermes Agent x Tvix

Persistent long-term memory for Hermes Agent, powered by Tvix. The plugin can recall both memories and Wiki documents, preserve conversation transcripts as Wiki source material, and mirror Hermes's built-in MEMORY.md into Tvix for cross-session persistence.

How It Works

Once installed, the plugin runs automatically during Hermes sessions:

Auto-Capture (on by default): Buffers user and assistant turns and writes original conversation transcripts to Tvix Wiki.
Auto-Recall (off by default): Searches past memories before the next response and injects relevant context.
Auto Wiki Recall (off by default): Searches wiki documents and injects stable reference material alongside memories.
Mirror Built-in (on by default): Mirrors Hermes built-in memory writes from MEMORY.md into Tvix in the background.
Knowledge Graph: Tvix organizes entities, relationships, and facts alongside vector search for richer retrieval.

Setup

Install the package and plugin

```bash theme={null}
uv tool install hermes-tvix && hermes-tvix install
```

Requires **Python 3.11+**.

If you prefer `pip`, `pip install hermes-tvix && hermes-tvix install` also works.

The install command copies the plugin into `~/.hermes/plugins/tvix/`, sets `memory.provider: tvix` in `~/.hermes/config.yaml`, writes default config to `~/.hermes/tvix.json`, and opens a browser for OAuth login.

Log in later if you skipped OAuth

```bash theme={null}
hermes-tvix login
```

Tokens are stored automatically in `~/.hermes/credentials/tvix.json`, so there are no API keys to manage manually.

Start using Tvix in Hermes

```bash theme={null}
hermes
```

Conversation transcripts are captured to Wiki automatically. To inject memories or Wiki docs before each response, enable `autoRecall` and/or `autoWikiRecall` in `~/.hermes/tvix.json`.

AI Tools

The AI uses these tools autonomously during conversations. These tvix_* tools are Hermes plugin-side wrappers around the corresponding Tvix MCP tools.

Tool	Wraps	Description
`tvix_search`	`search_memory`	Search memories by semantic similarity. Supports date filtering (`date_from`, `date_to`, `timezone`) and source filtering (`sources`). Defaults to 20 results and caps at 30. Returns a compact OpenClaw-compatible text list with related facts.
`tvix_store`	`add_memory`	Save important conversational context to long-term memory. Useful for preferences, goals, decisions, and project context. Requires a display summary and supports content up to 50,000 characters.
`tvix_forget`	`delete_memory`	Delete a memory. Shows candidate matches first, then deletes after confirmation.
`tvix_profile`	`tvix://profile`	Retrieve the user profile and related memories for session context.
`tvix_search_wiki`	`search_wiki`	Search Wiki documents for stable factual references. Defaults to 10 results and caps at 20. Results label Project, Basic, or Unknown destinations.
`tvix_add_wiki`	`add_wiki`	Create a Wiki document from full markdown content and an optional Project filing location. Reports the returned destination.
`tvix_update_wiki`	`update_wiki`	Update the title, content, or Project of an existing Wiki document. Set Project to `null` to move a document to Basic.
`tvix_delete_wiki`	`delete_wiki`	Delete a Wiki document. Shows candidate matches first, then deletes after confirmation.

Info: Hermes supports both memory tools and Knowledge Wiki tools, and also mirrors Hermes built-in MEMORY.md into Tvix for cross-session persistence.

CLI Commands

hermes-tvix install              # One-shot install and OAuth login
hermes-tvix login                # OAuth login (PKCE)
hermes-tvix logout               # Remove stored tokens
hermes-tvix status               # Check API connectivity and profile
hermes-tvix resync               # Rebuild mirror index from MEMORY.md
hermes-tvix resync --dry-run     # Preview mirror resync without writing

After installation, these commands are also available inside Hermes as hermes tvix <command>.

Auto-Recall

When enabled, Hermes prepares relevant Memory and Wiki context before the next AI turn and injects the result into the conversation.

Casual chat and short operational messages are skipped to keep recall focused. Both recall modes are disabled by default:

{
  "autoRecall": true,
  "autoWikiRecall": true
}

Auto-Capture

Auto-Capture buffers user and assistant messages during the session and stores original transcripts as Wiki documents.

During an active session, buffers flush after 5 minutes of silence or 20 buffered messages, whichever comes first.
At least 50+ characters are required before content is captured, which avoids storing tiny one-off messages.
Only the primary Hermes agent context is captured, so subagents and background contexts do not create noisy Wiki transcripts.

Mirror Built-in Memory

When Hermes writes to its built-in MEMORY.md, the plugin mirrors those writes to Tvix in the background. A local mirror_index.json file prevents duplicate uploads.

If you need to rebuild the mirror index from an existing MEMORY.md, run:

hermes-tvix resync

Configuration

Main plugin config is stored in ~/.hermes/tvix.json. Hermes itself is pointed at Tvix through ~/.hermes/config.yaml:

memory:
  provider: tvix

All keys in ~/.hermes/tvix.json are optional:

Key	Type	Default	Description
`apiUrl`	string	`https://api.membase.so`	Tvix API URL. Override only if instructed by Tvix support.
`tokenFile`	string	`~/.hermes/credentials/tvix.json`	OAuth token cache path. Stored outside the plugin directory so it survives updates.
`autoRecall`	boolean	`false`	Inject relevant memories before each response.
`autoWikiRecall`	boolean	`false`	Inject relevant wiki documents before each response.
`autoCapture`	boolean	`true`	Automatically store user/assistant conversation transcripts to Wiki.
`mirrorBuiltin`	boolean	`true`	Mirror Hermes built-in memory writes into Tvix.
`maxRecallChars`	number	`4000`	Max characters of recalled context per turn (500–16000).
`debug`	boolean	`false`	Enable verbose debug logging.

OAuth login stores rotating tokens in tokenFile and also persists the generated OAuth clientId automatically, so you normally do not need to edit auth-related keys by hand.

Example config:

{
  "autoRecall": true,
  "autoWikiRecall": true,
  "autoCapture": true,
  "mirrorBuiltin": true,
  "maxRecallChars": 4000
}

Updating the Plugin

To upgrade to the latest version:

uv tool upgrade hermes-tvix
hermes-tvix install --skip-login

If you installed with pip, pip install --upgrade hermes-tvix also works. If you want to re-authenticate as part of the upgrade, rerun hermes-tvix install without --skip-login.

How Tvix Differs

	Simple vector memory	Tvix
Storage	Flat embeddings	Hybrid: vector embeddings + knowledge graph
Search	Vector similarity only	Vector + graph traversal (entities, relationships, facts)
Extraction	Store raw text	AI-powered entity/relationship extraction
Knowledge base	Usually memory-only	Memory + wiki documents in one system
Auth	API key	OAuth 2.0 with PKCE (no secrets to manage)
Ingest	Synchronous	Asynchronous processing

GitHub Repository

Want to inspect the plugin source, track releases, or contribute improvements?

hermes-tvix on GitHub

Browse the source code, open issues, and follow plugin updates in the standalone repository.

Next Steps

Bring Your Context

Import chat history, connect apps, and more.

Use Your Context

Chat with Memory, agent retrieval, and dashboard exploration.

Knowledge Wiki

Store factual knowledge as wiki documents that Hermes can search directly.

Tvix MCP

See every memory and wiki tool available across clients.