my-project-public

repository

loading code, commits, and activity

repositories

loading repo index

#1	# Hermes Integration For Clawd Memory
#2
#3	Clawd Memory can run through the existing Hermes-compatible Mnemosyne provider. The public Clawd workflow is `clawd-brain` and `ClawdBrain`; the Hermes plugin keeps `mnemosyne` names because those are the current provider and tool contracts.
#4
#5	Use this integration when a Hermes agent should automatically recall and write Clawd memory during normal agent runs.
#6
#7	## Setup
#8
#9	### Step 1: Install
#10
#11	```bash
#12	pip install mnemosyne-memory
#13	```
#14
#15	Or from source:
#16
#17	```bash
#18	cd /Users/8bit/bots/Cladwbot-solana/solana-clawd/MemeBRain
#19	pip install -e ".[all,dev]"
#20	```
#21
#22	### Step 2: Register with Hermes
#23
#24	```bash
#25	python -m mnemosyne.install
#26	```
#27
#28	This creates a plugin entry at `~/.hermes/plugins/mnemosyne/` and wires up the MemoryProvider. That path is a compatibility name, not the product name.
#29
#30	### Step 3: Activate
#31
#32	```bash
#33	hermes memory setup
#34	# Select "mnemosyne" from the picker and press Enter
#35	```
#36
#37	### Step 4: Verify
#38
#39	```bash
#40	hermes memory status # Should show "Provider: mnemosyne"
#41	hermes mnemosyne stats # Working + episodic memory counts
#42	```
#43
#44	## How It Works
#45
#46	The provider hooks into the Hermes agent lifecycle:
#47
#48	\| Hook \| Behavior \|
#49	\|---\|---\|
#50	\| `pre_llm_call` \| Injects relevant working memory context into the prompt \|
#51	\| `on_session_start` \| Initializes session-scoped memory state \|
#52	\| `post_tool_call` \| Captures tool results as memories (if configured) \|
#53
#54	### Registered Tools
#55
#56	The provider registers these tools in the Hermes tool registry. Use them as Clawd memory tools inside agent runs:
#57
#58	\| Tool \| Description \|
#59	\|---\|---\|
#60	\| `mnemosyne_remember` \| Store a memory \|
#61	\| `mnemosyne_recall` \| Search memories \|
#62	\| `mnemosyne_stats` \| Show memory statistics \|
#63	\| `mnemosyne_triple_add` \| Add a knowledge graph triple \|
#64	\| `mnemosyne_triple_query` \| Query the knowledge graph \|
#65	\| `mnemosyne_sleep` \| Run consolidation \|
#66	\| `mnemosyne_scratchpad_write` \| Write to scratchpad \|
#67	\| `mnemosyne_scratchpad_read` \| Read scratchpad \|
#68	\| `mnemosyne_scratchpad_clear` \| Clear scratchpad \|
#69	\| `mnemosyne_update` \| Update a memory by ID \|
#70	\| `mnemosyne_forget` \| Delete a memory by ID \|
#71	\| `mnemosyne_invalidate` \| Mark a memory as superseded \|
#72	\| `mnemosyne_export` \| Export all memories to JSON \|
#73	\| `mnemosyne_import` \| Import memories from JSON \|
#74	\| `mnemosyne_diagnose` \| Run PII-safe diagnostics \|
#75
#76	## CLI Commands
#77
#78	```bash
#79	hermes mnemosyne stats # Current session stats
#80	hermes mnemosyne stats --global # Stats across all sessions
#81	hermes mnemosyne inspect "query" # Search memories
#82	hermes mnemosyne sleep # Run consolidation
#83	hermes mnemosyne export --output backup.json
#84	hermes mnemosyne import --input backup.json
#85
#86	# Import historical Hindsight memories via PR #28's timestamp-preserving importer
#87	hermes mnemosyne import --from hindsight --file hindsight-export.json --bank hermes
#88	hermes mnemosyne import --from hindsight --input hindsight-export.json --bank hermes
#89	hermes mnemosyne import --from hindsight --base-url http://localhost:8888 --bank hermes
#90
#91	hermes mnemosyne clear # Clear scratchpad
#92	hermes mnemosyne version # Show version
#93	```
#94
#95	## Data Location
#96
#97	By default, data is stored under:
#98
#99	```
#100	~/.hermes/mnemosyne/
#101	├── data/
#102	│ ├── mnemosyne.db # Main SQLite database (BEAM + legacy)
#103	│ ├── triples.db # Used by standalone TripleStore()
#104	│ └── banks/<name>/mnemosyne.db # Named memory banks
#105	└── ...
#106	```
#107
#108	This path is chosen because Hermes already persists `~/.hermes/` across sessions (including on ephemeral VMs like Fly.io).
#109
#110	## Auxiliary LLM routing (Codex / OAuth providers)
#111
#112	By default Mnemosyne uses its own LLM config (`MNEMOSYNE_LLM_BASE_URL` /
#113	`MNEMOSYNE_LLM_API_KEY`) or a local GGUF for sleep/consolidation and fact
#114	extraction. Hermes users with OAuth-backed providers like `openai-codex` can
#115	opt into routing those calls through Hermes' authenticated auxiliary client
#116	instead — no extra credentials required.
#117
#118	Set `MNEMOSYNE_HOST_LLM_ENABLED=true` to enable. See
#119	[hermes-llm-integration.md](hermes-llm-integration.md) for the full behavior
#120	model, configuration reference, and session-shutdown semantics.
#121
#122	## Optional MCP Server
#123
#124	For integration with MCP-compatible clients:
#125
#126	```bash
#127	mnemosyne mcp # stdio transport
#128	mnemosyne mcp --transport sse --port 8080 # SSE transport
#129	```
#130
#131	Mnemosyne does not currently expose a standalone REST API server.
#132
#133	## Uninstall
#134
#135	```bash
#136	python -m mnemosyne.install --uninstall
#137	hermes memory setup # Switch back to built-in memory
#138	```
#139

z6Mkq5mY3JWtxoxUobWcfNHm7AkRubgSWEZTkBVqZXJviFZ5/my-project-public