Search skills/

ai-memory agent skills

Self-hosted memory server and skills that let coding agents share a markdown wiki and hand off work between Claude Code, Codex and other CLIs.

akitaonrails/ai-memory 20 skills Memory & contextSkill & agent tooling

with caveats

A thorough, MIT-licensed memory system with evals and CI, but you are installing a server and hooks that record every session, not just a skill file.

check first: curl | sh installer · relies on an outside CLI → security details

Depth
4/5
Last update
today
Installs
—
Stars
15K
Forks
1.1K
License
MIT

What ai-memory does

ai-memory is a self-hosted memory server for coding agents, written in Rust, plus skills that teach the agent how to use its MCP tools. Lifecycle hooks record prompts, tool calls and session boundaries and sanitize them before storage. The server then consolidates that material into pages in a git-backed wiki of plain markdown files. The database is an index you can always rebuild from the files, and the default path makes no LLM calls. If you quit one agent and open another in the same directory, a handoff tells the next agent where you stopped, what failed and what is still open. Each handoff is single-use and can be claimed only once. Starting agents with ai-memory run codex also passes along a temporary local copy of the previous harness's visible history.

The skills are routing guides for groups of tools. ai-memory-handoff lists, accepts, begins and cancels handoffs. ai-memory-durable-pages writes, patches, deletes and restores wiki pages, and only when you ask. ai-memory-learning-maintenance covers consolidation, linting, forget sweeps and page feedback. The rest are retrieval, messaging, routing-install and status. The same set is duplicated under plugins/claude-code, plugins/codex and the core crate. Optional features include memory decay, compaction, a cross-project profile, multi-user auth with an audit log, and an LLM "dream" pass that is off by default.

Good for

  • Switching between Claude Code, Codex and other agents mid-task
  • Developers who want agent memory as editable markdown in git
  • Teams that want shared project memory on one self-hosted server
  • Resuming the same project across several machines

Not for

  • Anyone who wants a drop-in skill with no binary or server to run
  • Native Windows setups, which the README marks as experimental

What's inside depth 4/5

About seven distinct skills, duplicated across plugin folders to reach 20 and averaging 6.7 KB, with no scripts in the skill folders themselves; they front a full Rust server with CI and 47 eval files in the repo.

Skills
20
Reference docs
0
Scripts
0
Tests / evals
47
Templates
0
Avg SKILL.md
6.5 KB

Counted from the repository file tree. How we score depth.

Dependencies and costs

  • ai-memory binary and a running ai-memory server (native install, rpm, or Docker)
  • Rust 1.95+ only if building from source
  • Optional Anthropic or OpenAI API key for LLM consolidation and the dream pass
  • Native Windows is experimental; WSL2 is supported
  • Telemetry: none found. Hooks capture session activity, but it goes to the server you run. In ai-memory run mode the README says the remote server receives only the harness and session identity, not the local transcript.

Security review

Security: Check first

ai-memory is a memory server with agent plugins. Its hooks automatically capture your prompts and tool calls into an ai-memory server, and its skills read and write wiki pages, handoffs and cross-project messages there. Installing it means running the vendor's installer or binary and letting it write skill files into global agent folders. The Rust server code was not part of this review.

Worth knowing

  • Installs with curl | shThe install instructions pipe the latest release install.sh from the project's GitHub releases straight into sh.README.md:222
  • installs_hooksLifecycle hooks such as SessionStart run automatically, capture prompts and tool calls, and inject pending handoffs into the session; this is the plugin's documented core feature.plugins/claude-code/skills/ai-memory-handoff/SKILL.md:20
  • Relies on a CLI or package we could not reviewThe skills depend on the ai-memory binary and its MCP server, whose Rust source was not among the files provided for review.plugins/claude-code/skills/status/SKILL.md:13
  • Makes network requestsMemory tools talk to an ai-memory server, and optional answer synthesis and consolidation call a configured LLM provider.plugins/claude-code/skills/ai-memory-retrieval/SKILL.md:13
  • Changes files outside your projectGlobal installs write skill files under ~/.claude, ~/.agents, ~/.grok and ~/.copilot, plus ~/.grok/rules/ai-memory-handoff.md, and edit CLAUDE.md/AGENTS.md marker blocks.crates/ai-memory-core/src/routing_skills/ai-memory-routing-install/SKILL.md:32
  • Installs packages or toolsInstallation uses system packages (sudo dnf install of an RPM), a Docker container, or a downloaded installer and wrapper binary.README.md:290

What we checked

The repository (1325 files) on Nov 1, 2026. Not a full review: 1291 of 1325 files were not downloaded (large repository; we reviewed the skill folders); it relies on code outside the repository: Rust source of the ai-memory binary/server (crates/*, ~870 code files not provided); install.sh served from https://github.com/akitaonrails/ai-memory/releases/latest/download/install.sh; ai-memory-wrapper binary downloaded per README.md:360.

NVIDIA SkillSpector static scan: 17 pattern hits (Rogue Agent 5, Excessive Agency 5, Tool Misuse 3, Privilege Escalation 2, MCP Rug Pull 1, Supply Chain 1). The HIGH and LOW hits are the README's curl | sh installer. The sudo, systemctl, docker and ~/.local hits are install documentation. The 'run any command' hit is in the status skill, which in fact limits the agent to two ai-memory commands. The Grok rules write and the 'without asking' hit are documented install behavior and a quoted example preference, respectively.

A review, not a certification. Skills change after we check them — read the files before installing.

License

MIT. Commercial use allowed.

Can it run unattended?

Not really. It is a memory and handoff layer, not a task loop with stop conditions, though its handoffs and cut-off session detection help another agent resume after a long run stops.

Alternatives to ai-memory

memsearch

zilliztech/memsearch

Cross-agent persistent memory for Claude Code, Codex, OpenClaw, OpenCode and DeepSeek Harness: Markdown files as the source of truth, with a Milvus search index.

with caveats Security: Broad access depth3/5

memU

NevaMind-AI/memU

Memory layer that mines agent session logs into a shared markdown wiki and re-injects it across Codex, Claude Code, Cursor and others, locally or via memU Cloud.

with caveats Security: Broad access depth4/5

Obsidian Second Brain

eugeniughelbur/obsidian-second-brain

Persistent, self-rewriting memory in your Obsidian vault: 48 commands, a shared cross-agent logbook (Crew), scheduled maintenance and a research toolkit.

with caveats Security: Check first depth5/5

Continuous Claude

parcadei/Continuous-Claude-v3

A personal Claude Code infrastructure: ~160 skills, 32 agents and 30 hooks that inject skill hints, isolate subagent context to files, and fight compaction.

with caveats Security: Broad access depth3/5

FAQ

What is ai-memory for coding agents?

ai-memory is an open-source, self-hosted memory server for coding agents. It stores what your sessions learn as a git-backed markdown wiki and passes single-use handoffs between agents, so you can quit Claude Code and continue in Codex without re-explaining the work.

How do I install ai-memory?

On Linux and macOS the README gives two commands: a curl of install.sh from the GitHub releases piped to sh, then ai-memory setup. It also documents an rpm package, a Docker setup and a two-command Windows path. Claude Code and Codex plugins ship in the plugins/ folder.

Does ai-memory need an API key?

No. The README says capture, search, handoffs and memory decay all run with zero LLM calls. An Anthropic or OpenAI key is only needed for optional LLM features such as server-side consolidation and the dream pass.

Is ai-memory safe to install?

We found no hidden network calls or telemetry to the authors. The main things to know are the curl-to-sh installer, the fact that hooks record prompts and tool calls into your server, and a README note that LLM headers in the module settings end up in a world-readable file.

How is ai-memory different from Claude Code's built-in memory?

The README argues that built-in agent memory lives on one machine and belongs to one agent. ai-memory keeps memory in a server you run, shared across more than twenty harnesses, multiple machines and optionally a team, with the files as plain markdown you can grep or open in Obsidian.

Does ai-memory work on Windows?

WSL2 is listed as supported and native Windows as experimental. The README links docs/windows.md for WSL2, Docker Desktop and Windows service setups.

Curated by , AI enthusiast. Reviews are written by AI reviewers under our rubric and checked against the repository files.

Sources: GitHub repository (stars, license, file tree, last push Nov 1, 2026). Review based on the README and SKILL.md files as published.