Memory OS

v1.0.0

Persistent memory system for AI agents — daily logs, long-term memory, identity files, and heartbeat-driven recall. Solves context amnesia across sessions.

0· 331·4 current·4 all-time
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
VirusTotalVirusTotal
Benign
View report →
OpenClawOpenClaw
Suspicious
medium confidence
Purpose & Capability
Name, description, and requested resources align: the blueprint only creates local workspace files (SOUL.md, USER.md, MEMORY.md, daily notes, HEARTBEAT.md, etc.) and does not request binaries, credentials, or system access beyond a writable workspace. The files and platform overrides match the claimed purpose.
Instruction Scope
SKILL.md instructs only local file operations and a mandatory pre-install security audit, which is appropriate. However, AGENTS.md (included in the blueprint) advocates a startup behavior of 'Don't ask, just do it'—i.e., load memory files automatically each session. That gives the agent broad discretion to read persistent, potentially sensitive local files every session. While this is consistent with the memory purpose, it raises privacy/exfiltration risk because those files may be sent to remote LLM providers as part of normal prompt context.
Install Mechanism
Instruction-only skill; no install spec, no downloads, and no code files. This is low-risk from an installation/execution standpoint because nothing is written to disk by an external installer beyond what the agent will create at operator approval.
Credentials
The skill requests no environment variables, no credentials, and no special config paths. That is proportionate to a local file-based memory blueprint.
Persistence & Privilege
The skill does not set always:true and is user-invocable. Still, the included AGENTS.md encourages automatic per-session loading of the created memory files (no per-session confirmation). Autonomous invocation plus automatic loading of persistent files increases the chance that sensitive content will be included in prompts to remote models. The blueprint warns about privacy and advises not to load MEMORY.md in shared contexts, but enforcement is manual.
What to consider before installing
This blueprint is internally consistent with its goal of creating local memory files, but review before installing: - Read references/blueprint.yaml and references/guide.md fully (the SKILL.md itself advises this). The skill relies on you performing a manual security self-audit before any file creation. - Inspect every template (SOUL.md, USER.md, MEMORY.md, HEARTBEAT.md, daily note templates) for any secrets or PII and remove or redact anything sensitive. - Consider changing AGENTS.md startup behavior to require per-session confirmation before loading MEMORY.md or other sensitive files (i.e., disable the 'Don't ask, just do it' instruction if you want explicit consent each session). - Be aware that even though the blueprint makes no network calls itself, the platform will include file contents in model prompts during normal operation; if your provider or environment forwards prompts to external APIs, those files may be transmitted off-machine. If that is a concern, avoid storing highly sensitive data in these files or ensure you're in a private/offline execution environment. - Disable or carefully configure HEARTBEAT integrations (email, calendar, project checks) until you confirm they will not cause outbound network actions without explicit approval. If you want lower risk: install only SOUL.md/USER.md and keep MEMORY.md or daily notes private or gated behind an explicit 'load memory' command. If you need help making AGENTS.md require explicit confirmation each session, ask the skill author or your operator to modify the startup protocol before installing.

Like a lobster shell, security has layers — review code before you run it.

latestvk97bddsypjsgprpzbnbmyxbqmh82akh4

License

MIT-0
Free to use, modify, and redistribute. No attribution required.

Comments