Qoris Memory — Persistent Agent Memory

Persistent memory for OpenClaw agents via the Qoris MCP server. Explicit save/recall tools for cross-session context. User-owned API key, no automatic data c...

Security Scan

The skill is internally consistent with its stated purpose: it only requires a Qoris API key and workspace ID, documents that data is sent only on explicit tool calls, and contains no install actions or hidden endpoints in the provided materials.

Purpose & Capability

Name/description match requirements: the skill integrates with Qoris MCP and requires QORIS_API_KEY and QORIS_WORKSPACE_ID, which are appropriate and expected for a remote persistent-memory service.

Instruction Scope

SKILL.md instructs only how to configure the MCP server, set the two environment variables, and use explicit memory tools (save_memory, search_knowledge, etc.). It does not instruct reading unrelated files or exfiltrating data beyond the described MCP endpoint.

Install Mechanism

No install spec — instruction-only skill — so nothing will be downloaded or written by an installer. Repository/metadata files (package.json, mcp-config.json, README) are present but there is no automated install step in the skill definition.

Credentials

Only two environment variables are required (QORIS_API_KEY, QORIS_WORKSPACE_ID) and the primaryEnv is QORIS_API_KEY. These are proportional to a service that authenticates per-user and scopes data to a workspace.

Persistence & Privilege

always:false (good). The skill allows autonomous model invocation (default), meaning an agent with permission to call tools could call save_memory automatically; this is normal but worth noting because it determines when data is transmitted (SKILL.md claims transmissions only occur on explicit tool calls).

Assessment

This skill appears coherent and limited to storing only what you explicitly send to Qoris, but before installing: 1) Verify you trust https://qoris.ai and review their privacy/retention policies; 2) Create a restricted workspace and use a scoped API key if possible (so a compromised key has limited blast radius); 3) Verify the GitHub repository and confirm code matches these docs if you want deeper assurance; 4) Monitor audit logs (or enable Knox governance) to see memory read/write activity; 5) Be aware that while the skill claims 'no auto-capture', an agent that has permission to invoke tools could be induced to call save_memory — limit which agents/accounts can use the key and rotate keys if needed.

Purpose

This skill connects your OpenClaw agent to the Qoris MCP server so it can persist memories you explicitly save — across sessions, with version history and semantic search.

Important: memories are saved only when your agent calls save_memory (or you run a memory command). The skill does not monitor, capture, or silently persist conversation content. Nothing is transmitted until you make an explicit tool call.

Data Handling & Privacy

This skill sends data to an external service. Understand what before installing:

• What gets sent: only the exact content you pass to save_memory / search_knowledge / update_memory tool calls. Nothing else.
• Where it's stored: https://mcp.qoris.ai/mcp (TLS), scoped to your QORIS_WORKSPACE_ID. Your workspace is isolated from other users.
• Who can read it: only holders of your QORIS_API_KEY. The key is yours, generated by you at qoris.ai/dashboard.
• Retention: per the retention policy at https://qoris.ai/privacy. You can delete individual memories with delete_memory or rotate the API key in your dashboard to invalidate all access.
• No auto-capture: there is no automatic memory mode. Every persisted memory corresponds to an explicit tool call your agent made.
• Credential ownership: QORIS_API_KEY is your own per-user secret. No credentials are bundled with this skill. Treat the key as sensitive — create a restricted workspace for shared or enterprise use.

What Qoris Memory Does

Explicit cross-session save/recall

Your agent can persist information it decides is worth remembering via the save_memory tool, and recall it later via search_knowledge or get_memories. Memories survive across sessions and remain available until you delete them.

Versioned records

Every memory update is a versioned record with a timestamp. You can see the history of what was stored and when via get_document_full_content.

Memory branches

Create separate memory contexts (branches) for different projects, clients, or workspaces. Your agent operates in the right context for the right task.

Conflict surfacing

When multiple agents update the same memory, the MCP server surfaces contradictions for human resolution rather than silently overwriting.

Shared workspace memory

Memory is workspace-scoped. Agents sharing the same QORIS_WORKSPACE_ID share the same memory pool. Your team can coordinate through a shared knowledge base.

Semantic search

search_knowledge runs a semantic search across stored memories and returns cited answers grounded in what was actually stored — no hallucination.

Available Memory Tools

These are the only tools this skill exposes. Each is an explicit, user-triggered action — nothing runs automatically.

save_memory

Store a new memory with optional tags and metadata. Content is the exact string you pass.

get_memories

Retrieve memories the agent has previously stored, optionally filtered by tag or date.

search_knowledge

Semantic search across stored memories. Returns the most relevant records with their content.

update_memory

Update an existing memory record. Creates a new version, old version retained for audit.

delete_memory

Remove a memory. Deletion is logged in the audit trail.

list_knowledge_documents

List documents indexed in the knowledge base.

get_document_full_content

Retrieve the full content of a specific knowledge document.

Setup Instructions

Step 1 — Get your Qoris credentials

1. Create an account at qoris.ai
2. Navigate to your workspace dashboard
3. Copy your QORIS_API_KEY and QORIS_WORKSPACE_ID
4. Add them to your environment:

export QORIS_API_KEY="your-api-key-here"
export QORIS_WORKSPACE_ID="your-workspace-id-here"

Step 2 — Connect Qoris Memory MCP server

Add to your OpenClaw configuration:

{
  "mcpServers": {
    "qoris-memory": {
      "url": "https://mcp.qoris.ai/mcp",
      "headers": {
        "Authorization": "Bearer ${QORIS_API_KEY}",
        "X-Workspace-ID": "${QORIS_WORKSPACE_ID}"
      }
    }
  }
}

Step 3 — Verify memory is active

Start a new OpenClaw session and run:

/memory status

Memory + Knox Governance

Optional: pair with knox-governance for audit logging of every memory read and write:

clawhub install knox-governance
clawhub install qoris-memory-mcp

Constraints

Memory is workspace-scoped. Free tier includes up to 1,000 memories and 500MB knowledge storage. Paid plans unlock unlimited memories and storage.

Support and Documentation

• Documentation: https://docs.qoris.ai/memory
• Dashboard: https://qoris.ai/dashboard
• Privacy policy: https://qoris.ai/privacy
• Source: https://github.com/QORIS-AI/qoris-memory-openclaw-skill
• Support: [e-mail]

About Qoris AI

Qoris AI builds the trust and governance layer for enterprise AI agents. NVIDIA Inception Program member. Claude Partner Network member. Patent pending U.S. 63/907,730. Based in Stamford, CT.

qoris.ai