Skip to content

← agentcairn

agentcairn vs other agent-memory tools

Several tools now address local memory for coding agents. The useful distinction is not a benchmark headline; it is what stays durable, how context enters the system, which hosts get ambient behavior, and whether you want editable files or a managed database interface.

Last reviewed July 16, 2026 against each project’s public website, documentation, and repository. Product details change; follow the linked primary sources before making an architectural decision.

Direct local peers

Swipe to compare every tool →

DimensionagentcairnPMBcodemembasic-memory
Durable datacanonical Markdown vaultSQLite events + adjacent LanceDB vectorsSQLite + sqlite-vecMarkdown files
Human-readable sourceYes — Markdown edits feed recallDatabase; inspect through CLI/dashboardDatabase; inspect through CLI/viewerYes — two-way Markdown
Coding-agent integrationClaude Code, Codex, OpenCode, Cursor, Hermes, and moreClaude Code, Cursor, Codex, Zed, and MCP hostsOpenCode + Claude Code; Codex early betaClaude, Codex, Cursor, and other MCP clients; native Hermes plugin
Ambient capturehost hooks plus out-of-band transcript sweephooks journal work and inject recallplugins/hooks feed a shared event pipelineHermes: recall + per-turn/session capture; tool-driven through MCP elsewhere
Retrieval shapeBM25 + vectors + optional reranker + wikilinksBM25 + vectors + entity graph + optional rerankerSQLite FTS5 BM25 + sqlite-vec + rerankinghybrid full-text + vectors + Markdown knowledge graph
Open-source licenseApache-2.0Apache-2.0MITAGPL-3.0

The direct peers, fairly

PMB

A local-first coding-agent memory built around SQLite events, adjacent LanceDB vectors, MCP, ambient hooks, and a local graph/timeline dashboard. It emphasizes automatic prompt-time recall, fast asynchronous journaling, and measuring whether remembered lessons are followed. The durable interface is a portable database rather than editable Markdown. Official site

codemem

A local-first companion for agent-heavy engineering work. OpenCode and Claude Code are stable integrations; Codex is explicitly early beta. Plugins and hooks feed typed memories into SQLite, retrieval combines FTS5 BM25 with sqlite-vec, and an optional viewer and peer-to-peer sync make the database inspectable and portable across machines. Official site · Integration status

basic-memory

The closest data-model peer. Both products center readable Markdown and expose it through MCP. basic-memory also has a native Hermes plugin with search-before-answer recall and automatic capture, while its default search combines full-text and vectors. It is a broader human-and-AI knowledge system with two-way notes, semantic links, a local product, and optional hosted sync. agentcairn is narrower: coding-agent session lifecycles across its supported hosts, provenance-aware capture, temporal supersession, and a replaceable retrieval cache. Project and license · Hermes plugin · Search modes

Adjacent platforms

Mem0 is a configurable memory library, self-hosted service, and managed platform for LLM applications. Letta is a stateful-agent platform with its own coding agent and memory systems. Graphiti is an open-source temporal context-graph framework; Zep operates managed context-graph infrastructure at enterprise scale. These can be the right choice, but they are not drop-in equivalents to a local coding-agent memory companion.

Mem0 OSS · Letta · Zep and Graphiti

Choose by boundary

  • Choose agentcairn when the durable source should be an editable Markdown vault shared by supported coding-agent integrations.
  • Choose PMB when automatic coding-agent memory, a local database/dashboard, and lesson follow-through are the desired model.
  • Choose codemem when its OpenCode/Claude workflow, typed SQLite memory, viewer, and optional peer sync fit your stack.
  • Choose basic-memory when the broader product is a human-and-AI Markdown knowledge system with optional hosted sync.

How agent memory works · Bridge Claude Code memory · Inspect the vault in Obsidian

FAQ

What are the closest alternatives to agentcairn?

PMB and codemem are close product peers because they also provide local, cross-session memory inside coding-agent workflows. basic-memory is the closest data-model peer because it also centers human-editable Markdown and MCP.

Is agentcairn a Mem0 alternative?

For a personal coding-agent workflow, possibly. For an application that needs a memory library, self-hosted API, or managed platform, Mem0 is built for that broader deployment model and may be the better fit.

Can I move my memory off agentcairn?

Yes. The durable memory is plain Markdown in a folder you control. The DuckDB index is derived and can be deleted or rebuilt; it is not the source of truth.

Why are Mem0, Letta, and Zep not in the direct-peers table?

They solve adjacent but broader problems: application memory services, complete stateful-agent runtimes, and managed context-graph infrastructure. The table stays focused on tools that directly serve local coding-agent memory workflows.

Get started

Detect supported hosts first, then configure them without a prior persistent install:

uvx --from agentcairn cairn install # detect + preview; writes nothing uvx --from agentcairn cairn install --all # configure detected hosts

Swipe to view the full command →

GitHub · Full quickstart