EngramPort

Give any AI agent persistent memory. MCP-native, bring-your-own-LLM (any model via OpenRouter), graph-RAG.

engramport.com · Docs · Pricing · npm

EngramPort is the persistent-memory layer for AI agents. Your bot remembers across sessions, recalls by meaning, and synthesizes higher-order insights from the patterns in what it has seen. Built on the Model Context Protocol, it plugs into Claude Desktop, Cursor, Cline, the OpenAI Agents SDK, or any MCP-aware client with three lines of config.

npm install -g engramport

// claude_desktop_config.json
{
  "mcpServers": {
    "engramport": {
      "command": "npx",
      "args": ["engramport"],
      "env": {
        "ENGRAMPORT_API_KEY": "ek_bot_...",
        "ENGRAMPORT_NAMESPACE": "my-brain",
        "LLM_PROVIDER": "anthropic",
        "LLM_API_KEY": "sk-ant-..."
      }
    }
  }
}

Restart your client. Your agent now has memory.

What you get

Seven MCP tools, mapped 1:1 to a graph-RAG memory substrate:

| Tool | What it does | |---|---| | remember | Store a memory. Auto-links to similar memories already in your namespace. | | recall | Semantic search. Returns top-k matches with similarity scores and graph-expanded context. | | chat | Ask your brain a grounded question. Modes: reflex (fast), deep_think (multi-step plan + multi-recall + synthesis), intense (deepest pass, more queries, wider recall). | | upload | Ingest a document. Auto-chunked into linked memories. | | groom | Auto-discover typed edges between memories: supports, contradicts, synthesizes, and more. | | dream | Cluster analysis. Brain reads connected memories and produces higher-order INSIGHT and PRINCIPLE nodes. | | inspect | Brain vitals: memory count, edge count, Graph Quality Index. |

Bring your own LLM

EngramPort does not resell LLM calls. You pay for vector storage, embeddings, and the MCP transport on a flat tier; the LLM bill goes to your provider directly. Three providers natively, plus virtually any model (DeepSeek, Mistral, Llama, Qwen, and more) through OpenRouter:

| Provider | Set LLM_PROVIDER to | Suggested fast / balanced / intense models | |---|---|---| | Anthropic | anthropic | claude-haiku-4-5-20251001 / claude-sonnet-4-6 / claude-opus-4-7 | | OpenAI | openai | gpt-4.1-nano / gpt-4.1-mini / gpt-4.1 | | Google | google | gemini-2.0-flash / gemini-1.5-pro / gemini-1.5-pro | | OpenRouter | openrouter | any OpenRouter slug, e.g. openai/gpt-4o-mini / openai/gpt-4o / anthropic/claude-3.5-sonnet |

Defaults flip automatically based on your LLM_API_KEY prefix. Override any tier via env (FAST_MODEL, BALANCED_MODEL, INTENSE_MODEL) or in your dashboard at engramport.com/dashboard.

Why graph, not just vectors

Most memory layers stop at vector similarity. EngramPort builds a typed graph on top: every memory you store is auto-linked to its semantic neighbors, and the groom and dream passes promote dense clusters into named insights and principles. Recall returns direct vector matches AND graph-expanded context. The result is responses grounded in the patterns across your memories, not just the nearest paragraph.

The underlying substrate is Eidetic, a graph-RAG engine running on Pinecone for vectors and Supabase Postgres for the graph layer. EngramPort is the MCP wrapper that exposes the substrate to any agent.

Quickstart

1. Sign up at engramport.com/signup. Magic-link email. Choose a namespace. Paste your LLM provider key.
2. Get your config snippet. EngramPort issues you an ek_bot_... API key and a copy-paste config block.
3. Drop into your client. Paste into claude_desktop_config.json (or the equivalent for Cursor, Cline, your custom agent). Restart.
4. Use it. Ask your agent to remember, recall, or chat with grounded context.

See engramport.com/docs/quickstart for a step-by-step walkthrough.

How it fits with other agent frameworks

| Client | Config path | Notes | |---|---|---| | Claude Desktop | %APPDATA%\Claude\claude_desktop_config.json (Windows) or ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) | Restart the app fully after editing. | | Cursor | ~/.cursor/mcp.json or workspace .cursor/mcp.json | Cursor will surface tools in the chat sidebar. | | Cline | VS Code settings: cline.mcpServers | Same JSON shape. | | OpenAI Agents SDK | Pass to the agent's MCP server list | See docs/openai-sdk. | | Any MCP-aware client | stdio command + env | EngramPort speaks vanilla MCP stdio. |

What we provide vs. what you provide

| Layer | Provider | |---|---| | Vector storage (Pinecone) | EngramPort | | Graph database (Supabase Postgres) | EngramPort | | Embedding generation (OpenAI text-embedding-3-small) | EngramPort | | MCP transport (stdio + HTTP) | EngramPort | | Memory substrate (MandelDB) | EngramPort | | LLM completions (chat, dream synthesis) | You (your provider API key) |

Your LLM bill goes to your provider directly. Our infrastructure cost is covered by your tier subscription.

Pricing

| Tier | Free | Hobbyist | Pro | Team | Enterprise | |---|---|---|---|---|---| | Price | $0 | $9/mo | $29/mo | $99/mo | Contact us | | Namespaces | 1 | 10 | unlimited | unlimited | custom | | Memories stored | 1,000 | 100,000 | 1,000,000 | 1M per user | custom | | Daily groom/dream | 1/day | 4/day | 24/day | 24/day | custom | | intense mode | gated | yes | yes | yes | yes | | Aegis audit log | none | none | basic | full | full | | Workspace SSO | no | no | no | yes | yes | | Sentinel OS bundled | no | no | no | no | yes |

LLM cost is yours regardless of tier.

Sign up free at engramport.com.

Privacy and provenance

• Your memories are stored in your namespace on Pinecone and Supabase. We do not train models on your memories.
• Your LLM API key is encrypted at rest using AES-256-GCM. It is decrypted only at request time to forward your call to your provider.
• Every memory carries a cryptographic provenance hash via Aegis, a dual-strand DNA-seal of the content at creation time. Tampering is detectable.
• Full data export is available from your dashboard at any time.

See engramport.com/docs/privacy and engramport.com/terms.

Project

EngramPort is a product of Covenant Systems AI LLC, a North Carolina LLC. Source for this MCP wrapper is MIT-licensed. The underlying MandelDB substrate is hosted; reach us at [email protected] for enterprise self-host inquiries.

Contributing

Issues and pull requests welcome at github.com/covenantsystemsai/engramport. The MCP wrapper is intentionally thin; substrate-level changes live in the private MandelDB repo, but improvements to the wrapper, docs, and client-integration patterns are open contributions.

EngramPort, a Covenant Systems product · © 2026 Covenant Systems AI LLC · engramport.com