🌐 中文版: README.zh-CN.md
A cross-platform long-term memory Skill for OpenClaw, Hermes-Agent, Claude Code (Cursor), and OpenCode. Your AI remembers your preferences, research methods, coding style — and self-evolves.
📦 Repository: https://github.com/jasonatgit/echomind_memory.skill
| Feature | Description |
|---|---|
| Hermes Agent Memory Plugin | Implements Hermes Agent memory interface for automatic per-turn read/write. Code-driven, no LLM decision required, 100% reliable |
| Platform-Aware Memory | All contextual memories tagged with platform (hermes/openclaw/opencode); same platform weight unchanged, cross-platform reduced; user preferences isolated by platform |
| WAL Concurrent Mode | Supports multi-process concurrent read/write |
| Auto Migration | Automatic migration with legacy data isolation |
| Isolated User Preferences by Platform | Different platforms (OpenClaw, OpenCode, etc.) have independent memory |
New (MemoryProvider ABC compliance):
| Method | Description |
|---|---|
queue_prefetch() |
Compatible with Hermes v0.13.0+ new interface, eliminates per-turn AttributeError (was previously causing error logs on every turn) |
on_session_switch() |
Fixes session ID corruption after /resume /branch /reset operations |
on_pre_compress() |
Auto-saves memories before context compression discards them |
on_delegation() |
Captures sub-agent task experience into long-term memory |
Fixes:
| Item | Description |
|---|---|
agent_context Filter |
Auto-detects and skips cron / subagent / flush non-primary contexts to prevent memory contamination |
handle_tool_call JSON |
Tool call return value now uses JSON string format, compliant with Hermes ABC contract |
| Unified Write Guard | prefetch / sync_turn / on_session_end / on_memory_write now use unified skip_writes guard |
| Fix | Platforms |
|---|---|
main.py added call() dispatch function |
OpenClaw |
http_api.py retrieve/store endpoints pass platform param |
All |
code_format/cli.py fixed async→sync crash |
OpenCode |
skill.yaml added platform param + openclaw.call declaration |
OpenClaw |
- Platform-aware memory: same-platform weight ×1.0, cross-platform ×0.5
- Hermes Agent plugin: implements MemoryProvider interface, automatic per-turn read/write
- WAL concurrent mode + auto data migration
| Platform | Integration | Reliability |
|---|---|---|
| Hermes-Agent | Via call() generic interface |
★★★★★ 100% |
| OpenClaw | Via skill.yaml + main.py tool calls |
★★★★☆ LLM-decision |
| OpenCode (Devika / CodeAct) | Via CLI + JSON Schema standardized format | ★★★★☆ LLM-decision |
| Claude Code (Cursor) | Auto-write .echomind/ files, AI auto-reads context |
★★★★☆ LLM-decision |
| Capability | Description |
|---|---|
| 6 Memory Types | Context / Task / User / Knowledge / Experience / Research |
| RL Auto-Optimization | AI automatically adjusts memory weights based on positive/negative feedback — gets smarter with every use |
| Research Memory | Stores paper metadata, theoretical models, algorithms, and research notes |
| Code Style Memory | Records your type hint, comment, and function length preferences |
| Experience Reuse | Previously fixed bugs / used models → auto-suggested next time |
| Zero-Dependency Local Storage | SQLite persistence, no Docker / PostgreSQL / Redis required |
| Cross-Framework | LLM-independent; adapts to OpenClaw / Hermes / Claude Code / OpenCode |
Memory system optimized for Management Science & Engineering research. All domains customizable.
When queries touch these domains, research memory is automatically retrieved:
| Domain | Trigger Keywords |
|---|---|
| Operations Research | linear programming, integer programming, operations research |
| Supply Chain | supply chain, inventory, logistics |
| Decision Analysis | decision analysis, multi-criteria, AHP |
| Optimization | optimization, optimal, gradient |
| Simulation | simulation, Monte Carlo |
| Game Theory | game theory, Nash equilibrium |
| Forecasting | time series, forecasting |
| Project Management | critical path, project management |
| Queuing Theory | queuing |
| Feature | Description |
|---|---|
| Hermes Adapter Plugin | Implements Hermes memory interface automatic read and write each cycle. Code-driven, no LLM decision required, 100% reliable |
| Platform-aware Memory | All contextual memories are tagged with the platform (hermes/openclaw/opencode); same platform weight ×1.0, cross-platform ×0.5; user preferences isolated by platform |
| WAL Concurrent Mode | Supports multi-process concurrent read and write |
| Automatic Migration | Automatic migration |
| User Preferences Isolated by Platform | Different users, different applications, different platforms have independent preferences, isolating your memory |
In OpenClaw, Hermes-agent, or OpenCode, simply say or copy/paste:
Install EchoMind skills from: https://github.com/jasonatgit/echomind_memory.skill| Capability | Description |
|---|---|
| 6 Memory Types | Context / Task / User / Knowledge / Experience / Research |
| RL Auto-optimization | Weights auto-adjust based on positive/negative feedback |
| Research Memory | Paper metadata, models, methods, notes |
| Code Style Memory | Type hints, comment style, function length, project conventions |
| Experience Reuse | Previously fixed bugs / used models → auto-suggest next time |
| Zero-Dependency Storage | Pure SQLite, no Docker/PostgreSQL/Redis required |
| Cross-Framework | LLM-independent; works with any platform that supports HTTP or MCP |
Only 3 packages: pydantic + python-dotenv + numpy. SQLite is a built-in Python module.
When queries touch these domains, research memory is automatically retrieved:
Place the entire echomind_memory.skill/ folder into your skills/ directory — the framework will auto-load all tools.
The framework discovers tools via skill.yaml, then calls main.call(tool_name, **kwargs) for dispatch. No extra configuration needed.
Run the sync command in your project root:
# Install as MemoryProvider plugin
cp -r echomind_memory.skill ~/.hermes/plugins/echomind/
hermes config set memory.provider echomind
Or call via code:
```python
from main import call
call("sync_code_memory", project_root="/path/to/project", user_id="alice")Automatically generates two files for AI consumption:
.echomind/context.json: Structured preferences and experience.echomind/README.md: Human-readable summary
Get standardized JSON memory via CLI:
python -m example.opencode_call alice "supply chain coordination model"Output can be directly injected into LLM prompt:
cd ~/.openclaw/skills/echomind-memory && python3 main.py
cd ~/.opencode/skills/echomind-memory && python3 main.py
Service runs on `http://localhost:8005`. The LLM calls memory tools based on skill triggers.
## Quickstart
```python
import sys; sys.path.insert(0, '/path/to/echomind_memory.skill')
from core.memory_agent import MainMemoryAgent
# Initialize SQLite persistence (auto-creates ~/.echomind/memory.db)
init()
# Store memory
call("store_memory",
user_id="alice",
task_id="task-001",
context=[{"role": "user", "content": "What are common supply chain coordination models?"}],
task_status="completed",
success=True,
platform="hermes", # or "openclaw", "opencode"
)
# Retrieve memory
result = call("retrieve_memory", user_id="alice", query="supply chain coordination")
for m in result["working_memory"]:
print(f"[{m['source']}] {m['content'][:80]}")
# Record feedback (AI self-evolution)
call("record_feedback",
user_id="alice",
task_id="task-001",
feedback="positive",
retrieved_memories=result["retrieved_memories"],
)
agent.disable_persistence()
| File | Description |
|---|---|
example/hermes_call_example.py |
Full Hermes-Agent usage example |
example/openclaw_call.py |
Full OpenClaw usage example (with research papers) |
example/cursor_sync_example.py |
Claude Code / Cursor memory sync example |
example/opencode_call.py |
OpenCode CLI and API usage examples |
All persistent data stored in ~/.echomind/memory.db (SQLite file). Can be backed up or deleted at any time.
EchoMind Memory System (v1.0.10, Pure SQLite)
├── User Memory (preferences/habits/RL weights) → user_memory table
├── Task Memory (task status/steps) → task_memory table
├── Experience Memory (success/failure experiences) → experience_memory table
├── Context Memory (conversation context) → context_memory table
├── Knowledge Memory (domain knowledge) → knowledge_memory table
├── Research Memory (papers/notes) → research_papers + research_notes
└── RL Optimizer (feedback self-optimization, persistent weights)
AI is not a tool — it's a collaborator. A collaborator shouldn't have to "re-learn who you are" every session.
EchoMind gives your AI:
- Memory of your coding style, preferences, and habits
- Memory of bugs you've fixed and approaches you've tried
- Memory of your preference for semi-parametric models over covariance modeling
- Memory of that painful 3-hour debugging session → auto-avoid next time
- This isn't just a plugin. This is a multi-agent memory neural network for your AI.