MCP e chamadas de ferramentas

MemPalace

mempalace/mempalace

A local-first, open-source AI memory system that stores conversation history and project files as verbatim text for semantic retrieval, achieving 96.6% R@5 on LongMemEval without requiring LLMs or API calls.

Abrir repositório oficial ↗ Página inicial do projeto ↗
★ 58KEstrelas
⑂ 7,5KForks
675Problemas em aberto
PythonIdioma
MITLicença
Q@project.QualityScorePontuação editorial

Capturas de tela do projeto

Captura de tela de MemPalace Captura de tela de MemPalace

Visão geral

MemPalace provides structured, verbatim memory for AI workflows. It avoids summarization or paraphrase, storing original content in a structured index metaphor: people and projects are 'wings', topics are 'rooms', and original content lives in 'drawers'. This allows scoped semantic searches rather than querying a flat corpus. The system is local-first—nothing leaves the machine unless opted in—and features a pluggable retrieval backend, defaulting to ChromaDB. It also includes a temporal entity-relationship knowledge graph backed by SQLite and exposes 44 MCP tools for integration.

Principais recursos

  • Verbatim text storage without summarization or paraphrase
  • Structured memory index (Wings, Rooms, Drawers) for scoped search
  • 96.6% R@5 raw retrieval on LongMemEval without LLM/API calls
  • Pluggable storage backends (ChromaDB, SQLite, Milvus, Qdrant, pgvector)
  • 44 MCP tools for palace management, knowledge graph, and agent coordination
  • Temporal knowledge graph with validity windows backed by SQLite
  • Auto-save hooks for Claude Code, Codex CLI, and Cursor IDE
  • Local-first architecture with zero egress by default
  • Multi-arch Docker support (amd64 + arm64)

Requisitos, instalação e início rápido

It is recommended to install the CLI in an isolated environment using `uv tool install mempalace` or `pipx install mempalace` to avoid dependency conflicts. Alternatively, use `pip install mempalace` within an activated virtual environment. A Docker image is available for the MCP server or CLI: `docker pull ghcr.io/mempalace/mempalace:latest`. Data persists under the `/data` directory in the container.

Uso

Initialize a palace: `mempalace init ~/projects/myapp`. Mine content: `mempalace mine ~/projects/myapp` or `mempalace mine ~/.claude/projects/ --mode convos`. Search the palace: `mempalace search "why did we switch to GraphQL"`. Load context for a new session: `mempalace wake-up`. Run the MCP server via Docker: `docker run -i --rm -v mempalace-data:/data ghcr.io/mempalace/mempalace`.

Compatibilidade do modelo e casos de uso

Core retrieval and the hybrid pipeline require no LLM. The optional rerank pipeline (which promotes the best candidate out of the top-20) works with any capable model, verified with Claude Haiku, Claude Sonnet, and minimax-m2.7 via Ollama Cloud. Embedding models supported are `embeddinggemma-300m` (multilingual, ~300 MB) and `all-MiniLM-L6-v2` (English-only, ~30 MB).

Observações sobre licença e riscos

MIT license.

Editorial verification 2026-08-09: repository URL, owner, description, license and repository statistics were reviewed. License metadata: MIT. README was fetched for the channel draft; re-check repository dependencies, releases and model terms before production use.

Lançamento e manutenção

Not stated in the repository metadata

QwenPaw

agentscope-ai/qwenpaw

★ 33,9KPython

Composio

composiohq/composio

★ 29,5KTypeScript