Perenna

A lightweight, Git-backed permanent memory for AI agents. Claude Code, Codex,ChatGPT, Cursor, and other MCP clients can share durable memories withoutsharing a vendor account or conversation history.

• Separate MCP tools for reading, writing, and deleting memories
• Local stdio and single-user OAuth-protected Streamable HTTP transports
• Human-readable Markdown stored in an independent Git repository
• Local Vexor retrieval index that can always be rebuilt from Git
• Cross-process locking for multiple local agent processes

Why Perenna?

Your memory should follow you, not the agent you happen to be using.

Claude Code, Codex, Cursor, and ChatGPT keep memory in separate silos. Switchagents and your memory disappears. Switch machines and local memory staysbehind.

Perenna gives them one shared, Git-backed memory. Local agents and ChatGPTcan connect to the same self-hosted Perenna service, while every durable memorystays ordinary Markdown you can inspect, edit, version, and back up yourself.

Mem0's self-hosted stack is muchheavier, while its hosted Free Plan currently allowscustomer content to be used for model training and product improvement.

Perenna is different by design: no account, no proprietary memory cloud, nolock-in. Just your memories, in your Git repository, on infrastructure youcontrol.

Quickstart

Install with your AI agent

Paste this into Claude Code, Codex, ChatGPT Desktop, Cursor, or another codingagent with terminal and local MCP configuration access:

Install Perenna and connect it to this AI agent as a local stdio MCP server.
Work through the complete setup autonomously.

Use these as the source of truth:
- https://github.com/scarletkc/Perenna/blob/main/docs/getting-started.md
- https://github.com/scarletkc/Perenna/blob/main/docs/guides/client-setup.md

1. Detect the operating system, shell, and current MCP client.
2. Check for Python 3.12 or newer, Git, and uv. Install uv in user scope if it
   is missing. If Python or Git needs administrator approval, give me the exact
   command and stop there.
3. Install Perenna with `uv tool install perenna`. If Perenna is already
   installed, upgrade it with `uv tool upgrade perenna`.
   For Codex, also run `perenna skill install --agent codex`. For Claude Code,
   run `perenna skill install --agent claude-code`. Do not replace an existing
   modified copy or remove unrelated installed skills.
4. Check the effective Vexor embedding provider configuration. Reuse a working
   `~/.vexor/config.json` or inherited environment configuration. If none is
   available, ask me to choose between a remote provider and local embeddings.
   Explain that a remote provider receives memory text and search queries. For
   a remote provider, keep the provider and model in Vexor configuration and
   supply its secret through `VEXOR_API_KEY` or the provider-specific environment
   variable. For local embeddings, install `perenna[local]` and configure the
   local model according to the Perenna configuration reference. Verify the
   selected provider with `uvx vexor doctor` using the same environment that
   the Perenna process will inherit.
5. Ask whether I want to synchronize Perenna with a private Git repository. If
   I do, ask me to provide or approve its URL, run
   `perenna sync setup <repository-url>`, and verify it with
   `perenna sync status`. Treat repository creation, remote replacement, and
   reconciling diverged history as separate choices that require my explicit
   approval.
6. Register `perenna mcp --source <stable-client-name>` using the client-specific
   method in the setup guide. Preserve unrelated MCP servers and settings. Use
   a stable source such as `claude-code`, `codex`, or `cursor` for this client.
   For another client, use its official instructions for adding a local stdio
   MCP server. Make sure the Perenna process inherits `VEXOR_CONFIG_JSON`,
   `VEXOR_API_KEY`, or any provider-specific key used in step 4. Report only
   whether a secret is present.
7. Verify `perenna --help` and the saved MCP configuration. Reload MCP servers
   and call `memory_read` with `action: "list"` when the client supports it. If
   a restart is required, tell me the single restart step.
8. Report the commands run, files changed, and verification results. Keep API
   keys out of tracked configuration files.

Install a published release

Perenna requires Python 3.12+, Git, anduv.

uv tool install perenna

Install the optional memory behavior skill for the local client:

perenna skill install --agent codex
# or
perenna skill install --agent claude-code

Repeat --agent in one command when both clients should receive the skill.The configuration referencedocuments user and project scope, destinations, and replacement safeguards.

Codex and Claude Code can instead install the combined Skill and MCP connectionfrom Perenna's repository Marketplace. Follow thePlugin setup guideand choose one setup path per client.

Perenna needs a working Vexor embedding provider. For interactive providerselection and configuration, run:

uvx vexor init

Perenna automatically reuses ~/.vexor/config.json. When using process-levelconfiguration, make sure the MCP server receives VEXOR_CONFIG_JSON plusVEXOR_API_KEY or the selected provider's key from its host environment.Remote providers receive memory text and search queries.

If you choose local embeddings, also install Perenna's local extra:

uv tool install "perenna[local]"

Vexor provider configurationcovers remote and local setup. From the environment that starts the MCP client,verify the selected provider with:

uvx vexor doctor

Configure an MCP client to start:

perenna mcp --source <client-name>

Perenna creates its local data under ~/.perenna/ unless another home isconfigured.

To import, publish, or fast-forward compatible history through a private Gitrepository, run:

perenna sync setup <repository-url>

Install from source for development

git clone https://github.com/scarletkc/Perenna.git
cd Perenna
uv tool install .

Documentation

Start with thedocumentation index,then follow the path for your task:

• Getting started
• Client setup
• Plugin setup
• Self-hosting for ChatGPT
• Using permanent memory
• perenna-memory Agent Skill
• Configuration reference
• Architecture
• Development guide

License

MIT