local collaboration layer for AI coding agents
> /plugin marketplace add PCIRCLE-AI/memesh-llm-memory> /plugin install memesh@pcircle-memesh
Repo: PCIRCLE-AI/memesh-llm-memory
What's inside
π English | ηΉι«δΈζ | Deutsch
Each session, your AI coding agent starts from zero. It proposes the approach you rejected last month, trips over the same failing test, and asks you to explain the architecture it helped design.
MeMesh remembers for it. Claude Code hooks capture and restore routine context; supported clients share the same local SQLite database through their documented integration. It works with Claude Code, Codex, Cursor and other MCP clients.
you work with the agent
|
v
+------------------+ +------------------+
| capture | | recall |
| sessions, | ---> | at session |
| commits, fixes | | start and |
| (automatic) | | before edits |
+------------------+ +------------------+
| ^
v |
+----------------------------------------+
| ~/.memesh/knowledge-graph.db |
| decisions, lessons, links between them |
+----------------------------------------+
git commit, after a plan is approved or a question answered, when Claude stops, before context compaction, when you say "remember this" (5 languages), and before a risky command that repeats an accepted lesson. The plan/question and "remember this" hooks only remind the agent to call remember; the ninth command handles both Codex SessionStart and SessionEnd to register and retire an eligible ordinary Codex CLI session.http://localhost:3737/dashboard.| Platform | How | Notes |
|---|---|---|
| Claude Code | Plugin: hooks, MCP tools, /memesh skill | Full automatic capture and recall |
| Codex CLI | Plugin, or MCP server (memesh-mcp) | Zero-config plugin install, or codex mcp add memesh -- memesh-mcp |
| Gemini CLI | MCP server (memesh-mcp) | gemini mcp add -s user memesh memesh-mcp |
| Cursor, Cline and other MCP clients | MCP server (memesh-mcp) | Point the client at memesh-mcp |
| Hermes Agent | Native memory-provider plugin | docs/platforms/hermes-agent.md |
| OpenClaw | Native memory plugin | Source only; not published or live-tested: docs/platforms/openclaw.md |
| Your own scripts and apps | HTTP API from memesh serve | docs/platforms/universal.md |
| ChatGPT, Gemini web and other hosted chat | HTTP API through a local bridge you run | docs/platforms/README.md |
Claude Code's eight hooks provide automatic capture, recall, reminders, and safeguards. The Codex plugin wires its SessionStart integration and MCP tools automatically. MCP-only clients call recall and briefing themselves.
Recall and capture are local and deterministic: SQLite FTS5 search, explicit memory tools, and rule-based hooks. This version does not configure or call an LLM, embedding, or vector provider. Retired provider settings from older versions stay on disk but are ignored; memesh doctor names the top-level keys without reading or printing their values.
Plugin installs and the npm-global CLI share one database. Most Claude Code users want its plugin plus the CLI; Codex can use its own plugin or the CLI's MCP server.
Claude Code chat Terminal, Codex, Cursor
| |
v v
+-----------------+ +------------------+
| A: plugin | | B: npm global |
| /plugin install | | npm install -g |
| hooks + tools | | memesh CLI |
| + /memesh skill | | + memesh-mcp |
+-----------------+ +------------------+
| |
+---------------+------------------+
v
~/.memesh/knowledge-graph.db
(one file, both paths)
A. Inside Claude Code (hooks, tools and the /memesh skill are wired for you):
/plugin marketplace add PCIRCLE-AI/memesh
/plugin install memesh@pcircle-memesh
Restart Claude Code. A β MeMesh line appears at the top of the next session.
B. In a terminal (needs Node 22.13+):
npm install -g @pcircle/memesh
memesh doctor # checks local install health and prints fixes
memesh install-hooks # only if you skipped A: wires Claude Code, keeps your own hooks
For a zero-config Codex install, run codex plugin marketplace add PCIRCLE-AI/memesh and codex plugin add memesh@pcircle-memesh. The manual alternative is codex mcp add memesh -- memesh-mcp. For Cursor, add { "mcpServers": { "memesh": { "command": "memesh-mcp" } } } to ~/.cursor/mcp.json. The Dashboard's doctor banner can apply the two recoverable local repairs it knows how to verify; it never changes files just because the page was opened.
The plugin does not install the CLI. After
/plugin install, typingmemeshin a terminal sayscommand not founduntil you also runnpm install -g @pcircle/memesh. If you only use Claude Code chat, A alone is enough.
Update: Claude Code plugin: memesh upgrade-plugin (or npx @pcircle/memesh upgrade-plugin without the CLI). Codex plugin: codex plugin marketplace upgrade pcircle-memesh && codex plugin add memesh@pcircle-memesh. npm-global CLI: memesh update. Installing with an AI agent? Point it at llms-install.md.
memesh remember "Login uses OAuth 2.0 with PKCE"
memesh recall "login"
# -> finds the PKCE decision
memesh briefing # what the agent knows about this project
memesh serve # starts the local server and prints the dashboard URL
Keep memesh serve running and open the printed URL. In Claude Code you do not even need the terminal for memory tools: say "remember this" in chat, and once there is something to show, the briefing arrives on its own at session start.
Two things worth knowing once you have memories:
forget archives a memory instead of deleting it. A newer memory can replace an older one.work_package to prepare one calendar digest or bounded visible turns from the newest eligible recent Claude Code transcript. Transcript mode requires the client's single matching MCP file root; missing or ambiguous roots and bounded scan failures fail closed. Submission retains redacted source turns and only stages pending human review; agents cannot apply or reject it, and MeMesh calls no provider. The exact discovery bounds are in the API reference.Full command and tool reference: docs/api/API_REFERENCE.md. How it is built: docs/ARCHITECTURE.md. Contributing: CONTRIBUTING.md.
| Tool | What it does |
|---|---|
work_package | Prepare one bounded untrusted calendar digest or Claude Code transcript package under one matching MCP workspace root; submit one strict result for pending human review, or defer without durable change. Transcript submission retains bounded redacted source turns; no file path, hidden reasoning, provider, embedding, or vector data is exposed. |
remember | Store knowledge as observations, relations and tags β or pass free text as note and the title, observations and name are derived; replace corrects a memory in place |
recall | Local FTS5 search with multi-factor scoring (relevance, recency, frequency, confidence, recall impact) |
forget | Soft-archive (never deletes) or remove specific observations |
export | Back up, migrate, or move memories as JSON between compatible agents |
import | Import memories with merge strategies (skip / overwrite / append) |
learn | Record structured lessons from mistakes (error, root cause, fix, prevention) |
task_state | Read or record where the work stands β goal, next step, blocker, what was just finished |
briefing | The assembled work topology for any MCP client β this project's decisions, lessons, knowledge and recent activity by default (minimal); standard adds the fresh task state and closes with a capped index of the project's durable memories, full adds other projects and global memory (the briefing setting β minimal / standard / full); generic context stays quiet, while exact project + recipient can surface only that recipient's unfetched deliveries |
user_patterns | Analyze your work patterns β schedule, tools, strengths, learning areas |
improvement | Stage an evidence-linked product improvement for human review, or read its status; agents cannot accept or reject it |
message | Discover live agents, then exchange exact-recipient untrusted messages. Durable JSON payload max: 64 KiB; complete native envelope max: 16 KiB with distinct native_message_too_large and recipient_unavailable failures. Native acceptance, discovery, poll, and fetch never imply acknowledgement or disposition |
Scored Ranking β Results ranked by relevance (30%) + recency (25%) + frequency (18%) + confidence (17%) + recall impact (10%).
Agent messaging, the exact rules (full guide: docs/platforms/agent-messaging.md):
View @pcircle-aiβs profileContext compression plugin for Claude Code. Automatically trims large tool outputβJSON, YAML, stack traces, and logsβbefore it enters the context window. Works as a Claude Code plugin (automatic, zero-config) or as an MCP server (on-demand).
@pcircle-aiFAQ
memesh is a Claude Code plugin with 2 hand-picked skills for development work, indexed on Flowy. Install it with the command on its page. It includes memesh-review, memesh. Its skills do not fire on their own yet. Request auto-invocation to have Flowy route them as you prompt. Free and open source.
Is this plugin yours?
Claim it with GitHubSubmit a pluginPromote it