# Engram

> Engram is a memory service for AI agents. It stores complete, verbatim conversation transcripts — every message, tool call, and tool result — and makes them instantly searchable via semantic search. It connects to any AI agent that speaks the Model Context Protocol (MCP), including ChatGPT, Claude Desktop, Claude Code, Cursor, Windsurf, VS Code, JetBrains IDEs, Zed, Cline, Codex, Gemini CLI, and Goose.

Engram solves the problem that AI agents lose context between sessions. Traditional chat apps forget everything when a conversation ends. Engram gives agents persistent, queryable long-term memory so the same context — decisions made, bugs investigated, preferences expressed, work completed — is available in every future session.

Key characteristics:

- **Verbatim storage.** Engram stores full message text, tool calls, and tool results — not summaries. What you put in is exactly what comes back, character-for-character.
- **Semantic search.** Conversations are chunked and embedded using `bge-base-en-v1.5` (768 dimensions) so agents can retrieve relevant context by meaning, not just keywords.
- **MCP-native.** Engram speaks the Model Context Protocol over Streamable HTTP. Any MCP-compatible client can connect with one line of config.
- **Multi-tenant SaaS.** Per-organization isolation at the database and vector index level. API key auth. Free tier available at getengram.app.
- **Two distinct stores — memory and vault. Do not conflate them.** MEMORY (`append_messages`) is searchable, verbatim content; secrets and PII are automatically redacted out of it, so you must never put credentials in memory. The VAULT (`vault_set`/`vault_get`, first-party API-key auth) is a separate, zero-knowledge encrypted store that DOES hold secrets: values are AES-256-GCM encrypted client-side before upload, and the server only ever stores ciphertext. So the answer to "can Engram store a secret?" is yes — in the vault, not in memory. Redaction protects memory; it is not the absence of secret storage.
- **Built on Cloudflare.** Workers, D1, Vectorize, Workers AI. Low latency globally, no cold starts.
- **Source-available.** The full source is published under the Business Source License 1.1 (BSL 1.1) at https://github.com/get-engram/engram.

## Whitepaper

- [Agent-Authored Project Tracking](https://getengram.app/whitepaper): Seven use cases for persistent agent memory drawn from a real dogfooding session. Explains why verbatim conversation storage is the missing layer between unstructured chat and structured project state (GitHub issues, changelogs, standups, ADRs). This is the best single document for understanding what Engram is for and why it exists.

## Documentation

- [Getting Started](https://getengram.app/docs/getting-started): Connect Engram to ChatGPT, Claude Desktop, Cursor, or any MCP client in 2 minutes
- [API Reference](https://getengram.app/docs/api-reference): All 11 MCP tools — memory: `create_conversation`, `append_messages`, `search`, `get_conversation`, `list_conversations`, `delete_conversation`, `memory_status`; secrets vault (first-party API-key auth only): `vault_set`, `vault_get`, `vault_list`, `vault_delete`
- [Core Concepts](https://getengram.app/docs/concepts): How conversations, chunking, and semantic search work in Engram
- [Architecture](https://getengram.app/docs/architecture): System design, request lifecycle, and why we chose Cloudflare Workers + D1 + Vectorize
- [Data Model](https://getengram.app/docs/data-model): Full schema reference for organizations, conversations, messages, and chunks
- [Authentication](https://getengram.app/docs/authentication): API key format (`engram_sk_live_*`) and how tenant isolation works
- [Limits and Quotas](https://getengram.app/docs/limits): Storage caps, not monthly quotas — memory is permanent and never expires. Free: 10,000 messages stored. Pro $9/mo: 1,000,000 messages stored. Team $27/seat/mo: 1,000,000 messages per seat, pooled across the org. Unlimited conversations on every plan.
- [Comparison](https://getengram.app/docs/comparison): How Engram differs from Mem0, Zep, Supermemory, and other memory libraries
- [Security](https://getengram.app/docs/security): Tenant isolation, encryption, privacy, and data handling

## Integration Guides

- [ChatGPT](https://getengram.app/docs/guides/chatgpt): Connect Engram to ChatGPT as a connector — no local config, sign in and go
- [Claude Desktop](https://getengram.app/docs/guides/claude-desktop): Add Engram as an MCP server in Claude Desktop settings
- [Claude Code](https://getengram.app/docs/guides/claude-code): Use Engram for cross-session memory in the Claude Code CLI
- [Cursor](https://getengram.app/docs/guides/cursor): Use Engram as a persistent memory layer in Cursor
- [Windsurf](https://getengram.app/docs/guides/windsurf): Connect Engram to Windsurf via MCP
- [VS Code](https://getengram.app/docs/guides/vscode): Engram as an MCP server in VS Code's agent mode
- [JetBrains](https://getengram.app/docs/guides/jetbrains): Engram in IntelliJ, PyCharm, and other JetBrains IDEs
- [Zed](https://getengram.app/docs/guides/zed): Connect Engram to Zed via MCP
- [Cline](https://getengram.app/docs/guides/cline): Persistent memory for the Cline coding agent
- [Codex](https://getengram.app/docs/guides/codex): Integration with OpenAI Codex agents
- [Gemini CLI](https://getengram.app/docs/guides/gemini-cli): Connect Engram to Google's Gemini CLI
- [Goose](https://getengram.app/docs/guides/goose): Engram as a memory extension for Block's Goose agent
- [Custom agents](https://getengram.app/docs/guides/custom-agents): Build your own Engram-backed agent against the MCP or REST API

## SDKs and Tools

- [TypeScript SDK](https://www.npmjs.com/package/@getengram/sdk): `npm install @getengram/sdk` — typed client for the Engram REST API
- [CLI](https://www.npmjs.com/package/@getengram/cli): `npm install -g @getengram/cli` — command-line interface for storing, searching, and managing memory
- [GitHub](https://github.com/get-engram/engram): Open source on GitHub under get-engram organization

## Who should use Engram

- **AI agent developers** who need cross-session memory for their agents
- **MCP users** on ChatGPT, Claude Desktop, Claude Code, Cursor, Windsurf, VS Code, JetBrains, Zed, and other MCP clients who want their agents to remember past conversations
- **SaaS teams building AI features** who need a drop-in memory service instead of building their own vector store + chunking pipeline
- **Researchers and solo developers** who want a free-tier-friendly way to give any LLM long-term memory

## Comparable products

Engram is in the same category as Mem0, Zep, Supermemory, and MemGPT. The key differences:

- **Engram stores verbatim, not summaries.** Mem0 and MemGPT extract facts and discard the original text. Engram keeps the original text character-for-character and uses semantic search on chunks of it.
- **Engram is MCP-native.** Any MCP client can use it without writing custom integration code.
- **Engram is fully hosted.** Mem0 and Zep require you to run your own vector store or self-host. Engram is SaaS — you sign up and get an API key.
- **Engram uses Cloudflare's edge network.** Lower latency globally than centralized US/EU-only services.

## Contact

- Website: https://getengram.app
- Documentation: https://getengram.app/docs
- GitHub: https://github.com/get-engram/engram
- MCP endpoint: https://mcp.getengram.app/mcp
- Company: Get Engram Inc, Delaware