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 →
| Dimension | agentcairn | PMB | codemem | basic-memory |
|---|---|---|---|---|
| Durable data | canonical Markdown vault | SQLite events + adjacent LanceDB vectors | SQLite + sqlite-vec | Markdown files |
| Human-readable source | Yes — Markdown edits feed recall | Database; inspect through CLI/dashboard | Database; inspect through CLI/viewer | Yes — two-way Markdown |
| Coding-agent integration | Claude Code, Codex, OpenCode, Cursor, Hermes, and more | Claude Code, Cursor, Codex, Zed, and MCP hosts | OpenCode + Claude Code; Codex early beta | Claude, Codex, Cursor, and other MCP clients; native Hermes plugin |
| Ambient capture | host hooks plus out-of-band transcript sweep | hooks journal work and inject recall | plugins/hooks feed a shared event pipeline | Hermes: recall + per-turn/session capture; tool-driven through MCP elsewhere |
| Retrieval shape | BM25 + vectors + optional reranker + wikilinks | BM25 + vectors + entity graph + optional reranker | SQLite FTS5 BM25 + sqlite-vec + reranking | hybrid full-text + vectors + Markdown knowledge graph |
| Open-source license | Apache-2.0 | Apache-2.0 | MIT | AGPL-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.
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 hostsSwipe to view the full command →