Shared memory layer for all your AI coding agents — elephants never forget
Project description
elephagent
One memory layer for all your AI coding agents — elephants never forget.
English | 中文
The Problem
You use Claude Code, Cursor, and Codex. Each stores project knowledge in a different place. Switch machines, add a teammate, or try a new agent — and you start from scratch.
| Without elephagent | With elephagent | |
|---|---|---|
| Memory location | Scattered across CLAUDE.md, .cursor/rules/, AGENTS.md |
One .agent/ directory, auto-synced to all platforms |
| Switch tools | Re-teach every agent from scratch | All agents share the same memory instantly |
| New teammate | Copy-paste tribal knowledge | git clone and everything is there |
| MCP servers | Configure separately in each tool | Register once, available everywhere |
How It Works
elephagent stores everything in one Git-synced .agent/ directory and renders the config files each tool already knows how to read. Your AI agents can also read and write memory directly via a built-in MCP server.
flowchart LR
subgraph repo["Your Repository"]
direction TB
src[".agent/\n─────────────\nmemory/\n decisions.md\n workflows.md\n pitfalls.md\ntools/\n registry.json"]
end
src -->|"elephagent build"| CLAUDE["CLAUDE.md\n(Claude Code)"]
src -->|"elephagent build"| AGENTS["AGENTS.md\n(Codex)"]
src -->|"elephagent build"| CURSOR[".cursor/rules/\n(Cursor)"]
src -->|"elephagent build"| MCP[".mcp.json\n(all clients)"]
CLAUDE --> cc["Claude Code"]
AGENTS --> codex["Codex"]
CURSOR --> cursor["Cursor"]
MCP --> cc
MCP --> codex
MCP --> cursor
cc -->|"/el-remember"| src
cursor -->|"/el-remember"| src
codex -->|"/el-remember"| src
Getting Started
1. Install
pip install elephagent
2. Talk to your AI agent (recommended)
Open your project in Claude Code, Cursor, or Codex and use these commands:
| Command | What happens |
|---|---|
/el-init-memory |
Sets up .agent/ and generates all platform files |
/el-remember <note> |
Saves a note to shared memory |
/el-import |
Imports existing memories and skills from other platforms |
/el-sync-memory |
Commits and pushes memory to Git |
/el-check-memory |
Runs a health check on the setup |
/el-handoff |
Summarizes current session before switching tools |
/el-add-skill |
Creates a new shared skill |
Cursor note: Open the project folder in Cursor — it auto-detects the
agent-memoryMCP server. Enable it in Settings → Cursor Settings → MCP if prompted.
Before using
/el-sync-memory: Create a repo at https://github.com/new (private recommended) and rungit remote add origin <url>in your project directory.
3. Or use the CLI
# Initialize in your project (auto-runs `git init` if needed)
elephagent init
# Add a memory note
elephagent remember "This repo uses pnpm. Redis is required for API tests."
# Import existing memories and skills from other platforms
elephagent import
# Verify the setup
elephagent doctor
# Before syncing, create a repo at https://github.com/new (private recommended),
# then add it as the remote:
# git remote add origin https://github.com/you/your-repo.git
# Commit and push memory to Git
elephagent sync -m "update memory"
Built-in Skills
elephagent ships seven skills that work across Claude Code, Cursor, and Codex.
| Skill | Trigger phrases | What it does |
|---|---|---|
/el-init-memory |
"init memory", "set up agent memory" | Bootstrap .agent/ and generate platform files |
/el-remember |
/el-remember <note> (slash command) |
Save a note from the conversation to shared memory |
/el-check-memory |
"check memory", "memory status", "doctor" | Health-check the memory setup |
/el-sync-memory |
"sync memory", "push memory" | Build → commit → push to Git |
/el-add-skill |
"add skill <name>" | Create a new shared skill |
/el-import |
"import memories", "import skills", "import from cursor" | Import existing memories and skills from other platforms |
/el-handoff |
"handoff", "switch to cursor", "save context" | Summarize current session to shared memory before switching tools |
CLI Reference
| Command | Description |
|---|---|
elephagent init |
Bootstrap .agent/ and generate all platform files |
elephagent remember "..." |
Append a note and rebuild |
elephagent build |
Regenerate all adapter files from .agent/ |
elephagent import |
Import memories and skills (supports --from, --path, skill names) |
elephagent doctor |
Check that everything is in sync |
elephagent sync -m "msg" |
Build → pull → commit → push |
elephagent tool list |
List registered MCP servers |
elephagent tool add <name> |
Register a new MCP server |
Importing from existing setups
Already have a CLAUDE.md, .cursor/rules/, or custom skills? Import them in one command:
# Auto-detect and import everything (memories + skills from all platforms)
elephagent import
# Import from a specific platform
elephagent import --from claude
elephagent import --from cursor
elephagent import --from codex
# Import specific skills by name (searches ~/.cursor/skills/ and ~/.claude/skills/)
elephagent import my-skill another-skill
# Import skills from a custom directory
elephagent import --path /path/to/skills
In auto mode, elephagent scans:
- Project-local files:
CLAUDE.md,.cursor/rules/,AGENTS.md - Global skill directories:
~/.cursor/skills/,~/.claude/skills/
Only hand-written content is imported — files generated by elephagent build are automatically skipped. Existing files in .agent/ are never overwritten.
Adding MCP tools
# stdio server
elephagent tool add context7 --command npx --arg -y --arg @upstash/context7-mcp
# HTTP server with token from env
elephagent tool add figma \
--url https://mcp.figma.com/mcp \
--bearer-token-env-var FIGMA_OAUTH_TOKEN
Built-in MCP Server
elephagent ships a small MCP server at .agent/tools/mcp_server.py that lets agents read and write shared memory directly through the MCP protocol.
| Tool | Description |
|---|---|
agent_memory_read |
Read one or all memory files |
agent_memory_search |
Search across all memory |
agent_memory_append |
Append a durable note |
agent_tool_registry |
Read the shared MCP registry |
Why Git?
- Memory travels with the repo, not the machine.
- Works in CI, on new laptops, with new teammates.
- Full history and diffs for every memory change.
- No third-party service required.
Security
Never commit secrets into .agent/. Use environment variable references instead:
elephagent tool add internal-api \
--url https://example.com/mcp \
--bearer-token-env-var INTERNAL_API_TOKEN
.agent/.gitignore excludes local scratch files and secret-looking filenames by default.
Roadmap
- Git-synced shared memory
- Auto-generated adapters for Claude Code, Cursor, Codex
- Built-in MCP server
- Shared MCP tool registry
- Built-in skills for Claude Code, Cursor, Codex
- Importers for existing Claude / Cursor / Codex memories and skills
- Python SDK (
import elephagent) - Memory compaction for large histories
-
pipx/ Homebrew packaging - GitHub Action for CI validation
Contributing
Issues and PRs are welcome. Before submitting, run:
elephagent build
elephagent doctor
python3 - <<'PY'
from pathlib import Path
for path in ["elephagent.py", ".agent/tools/mcp_server.py"]:
compile(Path(path).read_text(), path, "exec")
print(path, "ok")
PY
License
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file elephagent-0.3.2.tar.gz.
File metadata
- Download URL: elephagent-0.3.2.tar.gz
- Upload date:
- Size: 19.9 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.11.3
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
52c50f30c01c8abe2f8e2bcdde197acbc54c77d0b6b62e0a23636dbf23b2b51f
|
|
| MD5 |
154ff72b3f82ae717940cfc7bae83cbb
|
|
| BLAKE2b-256 |
309bb11f52833fdd207bf1872bf35552909172e4b08f8e20fead09c5686321c4
|
File details
Details for the file elephagent-0.3.2-py3-none-any.whl.
File metadata
- Download URL: elephagent-0.3.2-py3-none-any.whl
- Upload date:
- Size: 20.8 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.11.3
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
ba60b5ecd184ed0318d2db0611239aa5755574bc09bc03df2d1476a0997dcd81
|
|
| MD5 |
61ceb5c3f5a1e83f381e1402450982c1
|
|
| BLAKE2b-256 |
639bf3a1ac7d275e1059bf0402e883c3c9df8197f3a32e087feb7b7bf0e8afd3
|