Engram

• CollectionMemory & State
• Classificationagent-native
• LicenseMIT
• MCPAvailable
• Agent SkillAvailable
• InterfacesAgent Skill · MCP · CLI · STDIO · http
• Verified2026-10-13

• Website ↗
• Docs ↗
• Official repo ↗
• Full dossier on GitHub ↗

What it does

Engram is a single Go binary that gives coding agents a persistent SQLite + FTS5 memory. Official positioning: the agent decides what is worth remembering via mem_save; Engram stores and searches it. Docs: “Engram trusts the agent to decide what’s worth remembering — not a firehose of raw tool calls.”

Next session, the agent calls mem_search / mem_context (and plugins can inject prior session context). Progressive disclosure is mem_search → mem_timeline → mem_get_observation. Optional git-chunk sync and opt-in cloud replication leave local SQLite authoritative.

Distinct from Claude-Mem: Claude-Mem auto-captures a tool-call firehose and compresses it in a worker. Engram is agent-curated FTS5 — one binary, zero Node/Bun/Python/Chroma. Their COMPARISON.md is the official contrast (note: that file’s Claude-Mem AGPL-3.0 license row is stale; Claude-Mem relicensed to Apache-2.0 in v13.0.0).

Agent-facing interfaces

• Interaction pattern: CLI + MCP / plugin
• Catalog protocols: Agent Skill, MCP, CLI, STDIO, http
• MCP: ✅ Available — stdio MCP is the primary agent surface.
• Agent Skill: ✅ Published as a Claude Code marketplace plugin with a Memory Protocol skill (not npx skills add).
• Onboarding type: mcp

How agents onboard

Interaction pattern: CLI + MCP / plugin

brew install gentleman-programming/tap/engram

See the full onboarding section in the dossier.

Protocol surface

Verification

Last verified 2026-10-13.

Sources:

• https://api.github.com/repos/Gentleman-Programming/engram