# Engram

> Persistent, searchable memory for AI agents and teams. Engram carries durable knowledge across sessions, MCP tools, repositories, and teammates.

Last updated: 2026-10-23

## What Engram does

- Gives AI agents durable memory across sessions, tools, repositories, and teammates.
- Keeps private workspace and repository knowledge isolated from a source-aware public memory bank.
- Supports hierarchical umbrella workspaces for multi-repository team memory.
- Connects through Model Context Protocol (MCP); no application SDK is required.
- Helps agents recover after context compaction, recall prior decisions, reuse lessons from other repositories, store durable discoveries, and report whether recalled knowledge helped.
- Collects sparse agent feedback pulse (useful, noisy, stale, missing_context, too_long) to continuously adapt retrieval confidence.
- Works with Claude Code, Cursor, Codex, Gemini, Antigravity, Copilot, OpenCode, Windsurf, Devin, Cline, Roo Code, Kiro, Amazon Q Developer, Goose, Continue, and other MCP-compatible clients.

## Verified product facts

- MCP tools: 17.

- Public memory bank: 100,000 source-aware memories.
- Guided skills: 20.
- Supported website languages: 16.
- LoCoMo evidence-retrieval diagnostic (N=1,536): MRR 0.5427, R@50 93.29%, R@200 97.59%.
- Full evidence-bearing LoCoMo answer diagnostic (N=1,536): Engram top-50 89.65%, 5.41 points above the next-best measured system.
- Benchmark results are local, reproducible diagnostics, not official leaderboard scores. Retrieval and answer quality are reported separately.

## Try Engram on one real task

Use the npm/stdio path for an isolated trial without email, browser authorization or an API key. The first memory call creates the trial. This path does not send a verification email.

If the user has authorized installing tools, add this server through the client's MCP settings, preserving existing servers:

{
  "mcpServers": {
    "engram": {
      "command": "npx",
      "args": [
        "-y",
        "engram-tools@latest",
        "mcp"
      ]
    }
  }
}

Or preview the guided setup with npx -y engram-tools@latest --setup --dry-run, then run it without --dry-run only with permission to edit client configuration. Reconnect the client so it discovers the tools.

Start with three useful calls, using the actual repository and task:
1. engram_start: describe the real project and current task.
2. engram_store: save one confirmed project decision and its reason. Do not invent a fact for the demonstration or store secrets.
3. engram_recall: in the next session, ask for that decision and its reason. Check that the memory belongs to the right project before applying it.

Trial allowance: 7 days, 25 private memories, 20 calls/day, 100 total calls. No payment is needed to try. An existing connection keeps its current workspace; do not replace a user's credentials to manufacture a new trial.

Run npx -y engram-tools@latest connect later to verify ownership and keep these memories in the same workspace. A human authorizes paid upgrades.

For remote HTTP-only directories, https://api.engram.tools/mcp currently uses OAuth and email verification. The npm email-less trial does not remove that remote authorization step. A crawler fetching this page has not installed or used Engram.


## Core agent flow

1. Start a meaningful session with engram_start.
2. After context compaction or a resumed session, call engram_rehydrate.
3. Before a non-trivial decision, call engram_recall.
4. Before rebuilding a pattern that may exist in another repository, call engram_cross_pollinate.
5. Store durable discoveries with engram_store; never store secrets or raw customer data.
6. Report whether recalled knowledge helped with engram_report_outcome or provide a sparse feedback pulse via engram_reflect (useful, noisy, stale, missing_context, too_long).

## Canonical sources

- Home: https://engram.tools/en
- Documentation: https://engram.tools/en/documentation
- Agent guide: https://engram.tools/en/agent
- Machine-readable agent guide: https://engram.tools/agent.txt
- Human setup: https://engram.tools/en/human
- Human product deck: https://engram.tools/en/deck/human
- Agent pitch deck: https://engram.tools/en/deck/agent
- Benchmarks and limitations: https://engram.tools/en/benchmarks
- Real use cases: https://engram.tools/en/use-cases
- Shared memory for AI agents: https://engram.tools/en/ai-agent-memory
- MCP memory server: https://engram.tools/en/mcp-memory-server
- Context compaction recovery: https://engram.tools/en/context-compaction-recovery
- Benchmark methodology: https://engram.tools/en/benchmark-methodology
- Articles and research: https://engram.tools/en/insights
- How to evaluate AI agent memory: https://engram.tools/en/insights/ai-agent-memory-benchmarks
- Claude Code integration: https://engram.tools/en/integrations/claude-code
- Codex integration: https://engram.tools/en/integrations/codex
- Gemini integration: https://engram.tools/en/integrations/gemini
- OpenCode integration: https://engram.tools/en/integrations/opencode
- Mem0 comparison: https://engram.tools/en/compare/mem0
- LangMem comparison: https://engram.tools/en/compare/langmem
- AgentMemory comparison: https://engram.tools/en/compare/agentmemory
- Full machine-readable context: https://engram.tools/llms-full.txt
- Sitemap: https://engram.tools/sitemap.xml