Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Memoria — 4-Tier Cognitive Graph Memory MCP

    1

    Persistent 4-tier cognitive graph memory & 27-tool MCP server that recalls architecture decisions and cuts token usage by 70%.

    Free

    Security scanned
    Memoria — 4-Tier Cognitive Graph Memory MCP

    Memoria — 4-Tier Cognitive Graph Memory MCP

    Example session with this skill installed

    Load memory context for this repository and check how we handled webhook signature verification previously.

    • Read your context and instructions
    • Compiled the memoria

    Memoria Context Loaded (4 Memories, 612 Tokens)

    TierConfidenceMemory Summary
    Core0.98All inbound webhooks must verify X-Signature-256 using crypto.timingSafeEqual before parsing JSON.
    Episodic0.91Fixed replay attack vector in src/webhooks.ts by enforcing a 300-second timestamp window (commit a41bc09).

    Next steps

    1. Reuse verifyHmacSignature() from src/lib/security.ts.
    2. Encode the new endpoint outcome into Memoria once tests pass.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    AI coding agents forget everything the moment a session ends, forcing you to re-explain architecture rules, past bug fixes, and repository conventions over and over. Memoria solves agent amnesia with a 4-tier cognitive graph memory engine and 27-tool MCP server.

    What it does

    • Token-budgeted context assembly (build_context) loads the highest-relevance Semantic, Episodic, and Core memories at session start within your exact token budget.
    • Hybrid Graph + Vector + BM25 recall (recall_memories) traverses typed edges between past tasks, files, and decisions.
    • Codebase & Diff intelligence (search_code, explain_code, explain_diff, explain_error) answers structural questions without reading dozens of raw files.
    • Automatic consolidation (consolidate, reflect) merges duplicate episodic memories into durable semantic rules and flags contradictions.

    How it works

    1. Derive namespace from the current project path (e.g., -Users-dev-myproject).
    2. Load session context via build_context with a 4,096-token budget before making edits.
    3. Encode outcomes via encode_memory (episodic for completed tasks, semantic for architectural invariants).
    4. Consolidate periodically to prune stale entries and strengthen verified patterns.

    Why this beats prompting it yourself

    Static CLAUDE.md files grow bloated and burn context window on every turn. Memoria dynamically retrieves only the memories connected to your current task, cutting prompt token consumption by up to 70%.

    Requires the Memoria MCP server configured in your MCP client. Free tier includes 1,000 API calls/memories; local SQLite/graph mode is unlimited.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 4 days ago

    • Free with an Agensi account

    Requires

    Node.js

    Permissions

    Read Files
    Network Access
    Listed4 days ago

    What's inside

    Frequently Asked Questions

    Static rule files are injected in full on every turn regardless of relevance. Memoria stores tens of thousands of episodic and semantic memories in a graph and retrieves only the exact subgraph relevant to your current prompt within a strict token budget.

    Any agent that supports SKILL.md and the Model Context Protocol (MCP), including Claude Code, Cursor, OpenCode, Codex CLI, and Windsurf.

    Yes. Each repository uses its own isolated namespace derived from the working directory path, while global engineering rules can be recalled across namespaces.

    New in LLM Ops