← Back to Skills Marketplace
318
Downloads
2
Stars
0
Active Installs
26
Versions
Install in OpenClaw
/install openclaw-memoria
Description
The most advanced memory system for AI agents. 24 cognitive layers, knowledge graph, procedural learning, dialectic queries, AI self-observation, auto skill...
Usage Guidance
This plugin appears to implement a sophisticated local-first memory system and most of its dependencies (SQLite, Ollama, optional cloud fallbacks) match that purpose. However: 1) Do not run the curl | bash installer without inspecting it first — download the installer and read it line-by-line before execution. 2) Inspect auto-skill.ts and any code that writes to ~/.openclaw or modifies openclaw.json to confirm it only adds the plugin and does not alter unrelated plugins or system-wide configs. 3) Backup your OpenClaw workspace (memoria.db, cortex.db, facts.json, and openclaw.json) before enabling migration or automatic configuration. 4) If you are sensitive about data leakage, avoid supplying remote API keys (OPENAI_API_KEY, OPENROUTER_API_KEY) and prefer local models (Ollama/LM Studio). 5) Consider enabling the plugin in a test or isolated environment first (or a container/VM) to validate behavior and resource usage (model pulls, disk writes, DB migration) before deploying in production. If you want, I can fetch and highlight the specific lines in the installer or auto-skill code that write configs or create files so you can inspect them quickly.
Capability Analysis
Type: OpenClaw Skill
Name: openclaw-memoria
Version: 3.34.0
Memoria is a highly sophisticated and feature-rich memory system for AI agents, implementing 21 cognitive layers including knowledge graphs, procedural learning, and vector search. The bundle performs extensive workspace file operations (reading USER.md/COMPANY.md and writing to SQLite/Markdown) and communicates with LLM providers (Ollama, OpenAI, Anthropic), but all these behaviors are clearly documented in SKILL.md and aligned with the stated purpose of providing persistent agent memory. The code is well-structured, includes a detailed changelog, and lacks any indicators of malicious intent, unauthorized data exfiltration, or hidden execution logic.
Capability Assessment
Purpose & Capability
The name/description (local-first multi-layer memory) aligns with the code and declared providers: SQLite, Ollama and optional remote fallbacks (OpenAI/Anthropic/OpenRouter). However the feature list includes 'auto skill creation' and continuous real-time capture which are more powerful than a passive memory store — these are plausible for an advanced memory system, but they materially expand what the plugin can do (create/modify files, auto-configure OpenClaw). Review auto-skill.ts and any code that writes into the extensions/plugins directories.
Instruction Scope
SKILL.md and INSTALL.md instruct the operator to run a remote install script (curl | bash) and the plugin registers hooks that read conversation content and explicit workspace files (USER.md, COMPANY.md, projects/*), write memoria.db and optional markdown outputs, and migrate older DBs (cortex.db, facts.json). Those actions are consistent with the purpose but the install instructions and hooks give the plugin broad file access inside the OpenClaw workspace and can auto-configure openclaw.json. The README/SKILL.md warns about this, but the scope is significant and should be inspected.
Install Mechanism
INSTALL.md includes a curl -fsSL raw.githubusercontent.com ... | bash installer flow. raw.githubusercontent.com is a common host but piping a remote script to bash is high-risk — it runs arbitrary commands with the user's privileges. The rest of the install (pulling large local models via Ollama, cloning repo, npm install) is expected for a local LLM-first plugin, but the remote-script install path and its auto-configuration behavior raise risk and deserve manual review before use.
Credentials
Declared environment variables are optional and proportional: OPENAI_API_KEY and OPENROUTER_API_KEY as fallbacks make sense given the plugin's support for remote providers. OPENCLAW_WORKSPACE is runtime-provided. The plugin also reads files in the declared workspace (USER.md, COMPANY.md, projects/*) — that is justified by the 'identity-aware' features and is documented in SKILL.md/Security.md.
Persistence & Privilege
The skill does not request always:true, and uses standard plugin hooks (before_prompt_build, after_tool_call, agent_end, after_compaction). It writes a persistent DB (memoria.db) into the workspace and can auto-migrate older memory files. The 'auto skill creation' capability and the install script's auto-configure step mean the plugin may modify OpenClaw configuration and create files — expected for the stated functionality but worth auditing before granting the plugin persistent presence and autonomous invocation.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install openclaw-memoria - After installation, invoke the skill by name or use
/openclaw-memoria - Provide required inputs per the skill's parameter spec and get structured output
Version History
v3.34.0
Fix recall context bloat
v3.33.0
v3.29-3.33: Async prefetch, WAL crash recovery, AI self-observation, auto skill creation, dialectic memory. 24 cognitive layers.
v3.27.0
Major refactor: Core modules moved to /core for improved modularity and maintainability.
- Split all main logic into a new /core directory; bootstrap, config, database, providers, and memory layers are now separate modules.
- Rewrote imports and references to use new /core module structure.
- Removed legacy top-level files, moving or rewriting their logic under /core.
- Updated keywords and tags for broader compatibility (now references Claude, Cursor, Copilot, ChatGPT, etc.).
- Expanded documentation to reflect new modular design and enhanced cross-agent support.
v3.25.1
- Increased cognitive layers from 20 to 21 in description and metadata.
- Version bump to 3.25.1.
- Documentation updated to reflect new version and improved layer count.
v3.25.0
- Adds 8 new files, expanding modularity (capture, config, continuous, extraction, format, orchestrator, procedural-hooks, recall).
- Updates core logic across 12 files, including memory extraction, procedural hooks, LLM/provider integrations, and the main entrypoint.
- Increments persistent memory layers to 21, as reflected in the description.
- Improves plugin structure for continuous capture and memory recall capabilities.
- Updates documentation and metadata for new features and version alignment.
v3.24.0
- Added Typescript support (includes openclaw.d.ts and tsconfig.json).
- Expanded to 21 cognitive layers and enhanced knowledge graph features.
- Improved procedural learning, observation handling, and continuous memory capture.
- Several core files updated for better modularity and maintainability.
- Minor updates to documentation and skill metadata for clarity.
v3.23.0
Phase 0: align repo truth
v3.22.3
- Update version to 3.22.3
- Documentation improvements in SKILL.md
- No functional or breaking changes detected
v3.22.2
v3.22.2: Layer 21 Continuous Learning + full code audit + JSDoc + dev guide
v3.21.0
Deep audit: 10 bugs fixed — hebbian was dead, proactive revision never triggered, storeFact lost 6 columns
v3.17.0
**Big update: Major refactor with granular modules, enhanced security details, and explicit plugin entrypoint.**
- Split functionality across new modules: added 24 files for distinct memory features (budget, embeddings, graph, scoring, procedural, etc.)
- Added detailed security & privacy section to clarify data flow, file access, and network behavior.
- Documented all active OpenClaw hooks used by the plugin for transparency and easier maintenance.
- Declared explicit plugin entrypoint (`index.ts`) in metadata.
- Improved installation/configuration instructions and clarified environment variable usage in SKILL.md.
v3.16.0
- Added new environment variables: OPENAI_API_KEY and OPENCLAW_WORKSPACE for improved fallback support and workspace-specific storage.
- Documented a new security section clarifying local data storage and usage of API keys only if remote providers are enabled.
- Updated environment variable descriptions for clearer setup of LLM provider fallbacks.
- Bumped version to 3.16.0 in documentation.
v3.15.2
- Major upgrade: Expanded from basic persistent memory to a 7-layer, multi-provider system with per-layer LLM configuration.
- Added support for Ollama, LM Studio, and remote APIs (OpenAI, Anthropic, OpenRouter) with automatic fallback chains.
- Now includes local embeddings (nomic-embed-text via Ollama) for zero-cost, offline use.
- Supports per-layer LLM choice (different models/providers for facts, procedures, graph, etc.).
- Improved documentation to highlight differences from basic memory plugins and explain full configuration options.
- Expanded SKILL.md tags and updated metadata to reflect new features and supported platforms.
v3.15.1
- Bumped version to 3.15.1 in package.json.
- No other changes made to the code or documentation.
v3.15.0
- Updated SKILL.md with clearer description, feature highlights, installation and configuration instructions, and community feedback links.
- Enhanced tags to include lifecycle and error-detection features.
- Added optional environment variable (OPENROUTER_API_KEY) for remote LLM fallback support.
- Updated author and feedback contact information.
- No code or core logic changes included in this version.
v3.11.0
- Update peer dependency range for openclaw in package.json to allow newer versions.
- No functional changes to the skill itself.
v3.10.0
openclaw-memoria 3.10.0
- Removed legacy and redundant files: INSTALL.md, SECURITY.md, install.sh, and openclaw.plugin.json.
- Updated documentation and metadata to reflect recent changes and streamline maintenance.
- Minor adjustments to configuration files for improved clarity.
v3.7.3
- Removed legacy, benchmark, and provider files to streamline the codebase.
- Updated package information in SKILL.md and package.json.
- No changes to user-facing features or documentation.
v3.7.2
openclaw-memoria 3.7.2
- Updated documentation and metadata files for consistency.
- Minor internal improvements and maintenance.
- No breaking changes or new features added.
v3.7.1
- Patch release with minor improvements and fixes.
- Updated index.ts and procedural.ts for improved stability.
- Refreshed package dependencies in package.json.
- Updated documentation in CHANGELOG.md.
Metadata
Frequently Asked Questions
What is Memoria?
The most advanced memory system for AI agents. 24 cognitive layers, knowledge graph, procedural learning, dialectic queries, AI self-observation, auto skill... It is an AI Agent Skill for Claude Code / OpenClaw, with 318 downloads so far.
How do I install Memoria?
Run "/install openclaw-memoria" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Memoria free?
Yes, Memoria is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Memoria support?
Memoria is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Memoria?
It is built and maintained by @Nitix_ (@nieto42); the current version is v3.34.0.
More Skills