# Porrima

> Self-hosted persistent mindspace for local AI agents.

Porrima is a full-stack local AI agent framework with persistent memory, project context, agentic tool execution, web tools, speech, artifact rendering, image workflows, backup/restore, and hardware/model observability. It is designed for single-user deployment on consumer workstation or home-server hardware.

## Primary Pages

- [Homepage](https://porrima.cc/): Product overview, screenshots, model recommendations, hardware requirements, and FAQs.
- [Documentation](https://porrima.cc/docs/): Installation, model choices, design principles, technology stack, user guide, and future work.
- [Installation](https://porrima.cc/docs/install/): Stable human-readable install page with agent one-liners and the canonical prompt.
- [Install prompt](https://porrima.cc/install/agent-prompt.txt): Plain-text prompt intended to be piped into an existing coding agent for guided installation.
- [Repository](https://github.com/asa-degroff/porrima): Source code and issue tracker for Porrima.

## Agent Installation

The canonical machine-readable installation prompt is:

```sh
codex "$(curl -fsSL https://porrima.cc/install/agent-prompt.txt)"
```

Equivalent one-liners on the docs page are provided for Claude, OpenCode, and Pi.

Security order matters: remote access should not be opened to an unprotected first-run instance. The install prompt requires Cloudflare Access or equivalent temporary protection before the first owner passkey is registered on the final HTTPS hostname.

## Important Capabilities

- Persistent first-person memory with extraction, reranking, passive recall, and synthesis cycles.
- Project-scoped chats and memory with remote-host project support.
- Managed llama.cpp model slots for chat, extraction, title generation, reranking, and embeddings.
- Streaming chat UI with thinking traces, tool calls, generated artifacts, speech, image workflows, and observability surfaces.
- Backup and restore for chat history, memory databases, and embeddings.

## Safety Notes

Porrima is intentionally powerful and is designed to run autonomously with access comparable to a coding agent. It is not a sandbox. Public or remote deployments should require HTTPS, authentication, and passkeys before exposure.

## Machine-Readable Files

- [Full LLM context](https://porrima.cc/llms-full.txt): Expanded site context and install prompt.
- [robots.txt](https://porrima.cc/robots.txt): Crawler policy and sitemap discovery.
- [security.txt](https://porrima.cc/.well-known/security.txt): Vulnerability reporting contact.
- [Agent skills index](https://porrima.cc/.well-known/agent-skills/index.json): Discoverable agent guidance for installing Porrima.