- Home
- Skills
- Agents & Orchestration
- MemPalace — Persistent Memory for Coding Agents
More screenshots
Works with the AI Agents you already use
MemPalace — Persistent Memory for Coding Agents
A local SQLite index (FTS5 BM25 + dense-vector embeddings, fused with reciprocal rank fusion) makes that memory searchable by your agent in milliseconds, entirely offline.
Free
MemPalace — Persistent Memory for Coding Agents
Example session with this skill installed
Set up MemPalace in a new project folder and prove it works.
- Read your context and instructions
- Compiled the mempalace
The agent created a clean folder, ran preflight (Python 3.13, uv 0.12.3, git 2.50.1; MemPalace 0.2.1 already present as a uv tool), then:
mempalace init-workspace --agent --agent-flavor antigravity
→ db_initialized: true, docs_indexed: 2, vectors_indexed: 2
created MemPalace/essences/2026-11-01-welcome-to-mempalace.md,
memory.sqlite, docs/decisions/, setup scripts,
.agents/skills/memory-sync/SKILL.md,
.agents/skills/decision-making/SKILL.md, MEMPALACE_KICKOFF.md
mempalace doctor --agent
→ all_ok: true, issues: []
sqlite 3.47.1 with FTS5 ok, fastembed ok, essences: 1, tools: 1, stale: 0
mempalace ingest → wisdom_upserted: 1, docs_indexed: 2
mempalace embed → model BAAI/bge-small-en-v1.5, vectors_added: 0
(initialisation had already embedded the seed documents)
mempalace search --mode bm25 "welcome"
→ 1 result: "Welcome to MemPalace" (wisdom, active)
mempalace search --mode hybrid "welcome"
→ 2 results: the welcome essence first, then the registered setup tool
Total wall-clock time for initialisation plus all verification steps: about 16 seconds. The workspace is left as plain files the user owns — no account, no cloud, no background service.
Connects securely to your tools. The creator never sees your data.
About this skill
What it does
Coding agents forget everything between sessions. MemPalace fixes that with a persistent, project-scoped memory that lives in your own repository: lessons (essences), architecture decisions (ADRs), session logs, and a tool registry — all as plain Markdown/JSONL files you can read, diff, and commit. A local SQLite index (FTS5 BM25 + dense-vector embeddings, fused with reciprocal rank fusion) makes that memory searchable by your agent in milliseconds, entirely offline.
This pack contains two skills:
- mempalace-setup — an installer and verifier. It checks your environment (Python 3.13+, uv), installs the MemPalace CLI from its public GitHub repository with your confirmation, initialises your workspace for your agent (Claude Code, Cursor, Antigravity/Gemini, OpenCode, Hermes, or generic), and proves the result with MemPalace's built-in doctor and a live search check.
- memory-sync — the day-to-day protocol. At the end of a task your agent scores candidate lessons against a deterministic quality gate (S = 0.35B + 0.30R + 0.20P + 0.15I ≥ 3.2), proposes the ones that pass in a compact block, and writes only the ones you confirm. It also covers session logging, tool registration, and periodic hygiene sweeps.
Decision governance (ADRs) is built in
Setup also initialises a decision log: docs/decisions/ with the ADR template, plus a bundled decision-making skill alongside memory-sync. Architecture, interface, process, and tooling decisions are recorded as numbered ADRs with their context and consequences. Once an ADR is accepted it is never edited — it is replaced by a new ADR that supersedes it — so the reasoning behind the current design stays auditable instead of being silently rewritten. ADRs are indexed into the same searchable memory as essences, sessions, and tools.
How it works
MemPalace itself is a free, open-source (MIT) Python tool installed from github.com/aruruka/mempalace with uv. These skills teach your agent when and how to install, verify, and use it. The memory files and the SQLite index are created inside your project and stay there — they are the output of the skills. No API keys are needed; the embedding model (bge-small-en-v1.5) runs locally and is downloaded once from Hugging Face.
Where it stops
- The skills run inside one project workspace you name; nothing is synced to any cloud service, and there is no background daemon.
- Installing software always needs your confirmation; the skills never download and pipe scripts into a shell.
- MemPalace is a single-user, single-project memory. It is not a team knowledge base and does not replace your project's documentation — it indexes the knowledge your agent accumulates while working in it.
Compatibility: Requires Python 3.13+ and the uv package manager; Windows, macOS, or Linux. Works with any agent that can run terminal commands and follow SKILL.md instructions. Network needed once: github.com (install) and huggingface.co (embedding model). Fully offline afterwards.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.
Requires Python 3.13+ and uv. Initial one-time download of embedding model (BAAI/bge-small-en-v1.5) requires internet. Operates only within the local workspace scope.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 10 days ago
- Free with an Agensi account
- Accounts
GitHub
- Software
Python 3.13+
uv (Python tool runner)MemPalace CLI
Install the free MemPalace CLI with uv from github.com/aruruka/mempalace.
Permissions
Allowed Hosts
File Scopes
Files are written only in the buyer's own project workspace. Network access is needed once: github.com (install) and huggingface.co (one-time embedding model download). No API keys needed; fully offline afterwards.