Polaris

Your AI project memory for Omarchy

Chats, meetings, mails and documents in. A sourced memory per project out.

Install • What's included • Docs • Architecture • Privacy

You work on projects with ChatGPT and Claude, and the decisions, facts and open questions end up scattered across dozens of chats, documents and mails. Polaris collects them per project while you work and has the coding agent you already use write a short, sourced memory for each project: what is known, what was decided, what is still open. It lives as plain Markdown in your own project folders, readable by you, your editor and every AI agent on your machine.

_{Illustration based on a demo project (invented content, translated). The curator currently writes its field labels in German („Fakt“, „Konfidenz“); the real file is shown further below.}

No account, no cloud service, no database. The only thing that leaves your computer is what the curator hands to your own agent CLI (Codex, Claude Code, or any command, local models included). Every entry links back to the chat message or file it came from.

In daily use with ChatGPT and Claude, whose project chats are captured automatically. Any other assistant that organises work in projects (Gemini, Grok, Perplexity, …) can feed Polaris today through the inbox, as exported or pasted chats and documents. Automatic capture for another assistant means one more extension module plus its mapping in the bridge; contributions are welcome (see CONTRIBUTING.md).

Install

1. Get it and run the guided setup (asks a few questions, shows a summary, then sets up):

git clone https://github.com/mstio/polaris-connector ~/Projects/polaris-connector
~/Projects/polaris-connector/install.sh

2. Add the browser extension (optional, for automatic chat capture): install the signed Firefox extension from the latest release and switch sync on in a ChatGPT or Claude project.

3. Work as usual. Chats are captured, the curator updates each project's memory at night or at a click in the bar. polaris update keeps everything current.

Details, options and updating: docs/install.md.

What you get

• Chats archived as you go. A Firefox extension saves finished turns of ChatGPT and Claude project chats into the matching project folder. No copy and paste, nothing from chats outside a project. Other assistants: via the inbox for now.
• Documents and mails, too. Drop files or .eml mails (with attachments, Word, Excel, PowerPoint) into a project's inbox/.
• A memory per project. knowledge.md, decisions.md and open-loops.md: condensed, deduplicated, kept current, every entry with its source. Unclear cases wait for you; Polaris never guesses a project.
• One glance in your bar. A theme-aware Omarchy widget shows new material per project, starts curation, opens the knowledge files and tells you when something is broken: capture, curation, backup, disk, security.
• Cheap in the background. No polling daemons: systemd path units, timers and on-demand processes only.
• More than memory: health check, security radar, backups and a shared memory for your AI agents; see What's included.

 

_{Left: the Polaris widget with the health check (English UI, rendered from a mock-up of the real panel). Right: the omarchy-backup widget (mock-up with the real widget texts).}

What's included

Polaris is a small suite. At its heart is the project memory; around it are tools that capture your material automatically and keep your Omarchy machine and your AI setup healthy. Pick what you need.

The project memory

• Polaris core: never lose track of a project again. Each project gets three short files: what is known, what was decided, what is still open. Plain text in your own folders, so you can read, search, back up or hand them to any AI. More
• Curator: you no longer write the summary yourself. Once a night (or at a click) your own AI assistant reads what is new and updates the three files: no duplicates, outdated facts replaced, and every entry says where it came from and how certain it is. More

Getting material in

• Browser capture: no copy and paste. Your ChatGPT and Claude project chats are saved to the right project folder while you work. Older chats can be imported from the official ChatGPT export. More
• Inbox: everything else counts too. Drop documents, notes, meeting transcripts or mails (with attachments) into a project's inbox folder. More

Keeping an eye on things

• Curator widget: your projects at a glance in the Omarchy bar: what is new, what is waiting, one click to update the memory or open the files. More
• Health check: know it works without checking. One panel shows whether capture, curation, backups, security and updates are fine, and says plainly when something needs you. More
• Radar and security check: hear about it before it bites. Daily, cheap check for security fixes that affect your machine, plus Omarchy news and plugin updates before you change your setup. More

Protecting your setup

• Backup: a broken update or a new laptop is no longer a lost weekend. omarchy-backup saves your working Omarchy setup, notices what changed since, checks that the copy in your cloud folder is really usable, and restores it on a fresh install. More
• Machine memory: your AI agents stop forgetting. Claude Code, Codex and others share one notebook about your machine (what was set up, why, what to avoid), so you don't explain it again in every session. More

Extras

• Agent launcher and dock: start Claude Code, Codex or Agy from the bar, and an app dock for Omarchy. More

Core and curator are always installed, everything else is optional; see Architecture for how the parts fit together.

Current extension version: see extension/manifest.json. What changed between versions: CHANGELOG.md (German). The module documents under docs/ are in English; curator internals are in German.

Privacy and legal notice

Polaris is a tool, not a compliance solution. You are responsible for using it in line with the law that applies to you (in the EU: the GDPR) and with your providers' terms. Please check before you use it, especially at work.

• What it stores: copies of your project chats, documents and mails, and curated summaries that can attribute statements to named people. All of it stays in your own folders; delete a project folder and it is gone.
• What leaves your machine: whatever the curator hands to the agent CLI you configure. That provider's terms and data processing apply (for example OpenAI for Codex, Anthropic for Claude Code). To keep everything local, use a local model via runtime: command.
• Other people's data: chats and mails often contain names, contact details and opinions of third parties. Purely personal use may fall under the household exemption; professional use usually does not. Then you typically need a legal basis, a data processing agreement with the model provider, information for the people concerned and deletion rules. Ask your data protection officer or legal adviser.
• Provider terms: the extension reads your own chats through the web apps' internal interfaces. Check that this is compatible with the terms of the services you use.

This notice is not legal advice. Polaris comes without warranty (MIT licence).

Architecture: core and modules

Polaris is a small core with modules around it. The core is the project memory; everything else feeds it, shows it or keeps the machine healthy.

Part	What it is	Lives in
Core: project memory	Project folder layout, the curator that condenses sources into knowledge, the polaris / polaris-curator CLIs	curator/, bridge/polaris/mapping.py, bridge/bin/polaris
Source: browser capture	Firefox extension + native messaging host + bridge; archives ChatGPT and Claude project chats turn by turn into raw/	extension/, bridge/
Source: ChatGPT data export	Fills gaps from the official export, or files old chats as Markdown without curation	polaris import-export, polaris archive-export
Source: inbox	Documents, notes and mails dropped into <project>/inbox/	curator/ (see curator/README.md)
Model runtimes	The curator calls an agent CLI: codex, claude or any command	curator/curator.example-*.yaml
Display: curator widget	Omarchy bar widget: pending sources, start runs, open knowledge files, health check	omarchy-plugin/mst.polaris-curator/
Cross-cutting: health check	One overview of capture, curation, backup, memory, security and system state	curator/polaris_curator/systemhealth.py
Cross-cutting: radar	Omarchy news, plugin updates and a daily security check (Arch tracker, kernel, CISA KEV, NVD)	memory/tools/polaris-radar
Cross-cutting: machine memory	Shared Markdown memory for all local AI agents, with linter, template and mirror	memory/
Cross-cutting: backup	omarchy-backup: snapshots, baseline drift, verified remote backup, fresh-install recovery, plus its bar widget	separate repos, fetched by install.sh via git
Extras	Agent launcher widget and dock	separate repos, fetched by install.sh via git
Developer tooling	Privacy check with protected folders, template sync check, public mirror	scripts/

 ChatGPT / Claude (Firefox)     official export      documents, mails
          |                          |                      |
   extension -> native host     polaris import-export       |
          |      -> bridge           |                      |
          v                          v                      v
   <project>/raw/<chatgpt|claude>/  ...................  <project>/inbox/
                         \                              /
                          `----> polaris-curator <-----'      model runtime
                                       |                  <-- (codex, claude,
                                       v                       any command)
              <project>/knowledge.md, decisions.md, open-loops.md
                                       |
          curator widget, health check, your agents, your editor

Direction of travel: the health check should only read one state file per module (the radar already works that way with ~/.local/state/polaris/security.json), so that it can become a standalone package and modules can be installed one by one. Today install.sh still sets up all modules together and the health check knows each of them.

Documentation

Topic	Document
Install, options, updating, switching from the old bundle	docs/install.md
Project folders, mapping, curator	docs/core.md
Browser capture (extension, bridge, troubleshooting)	docs/browser-capture.md
ChatGPT data export	docs/chatgpt-export.md
Inbox (documents and mail)	docs/inbox.md
Model runtimes	docs/model-runtimes.md
Curator widget	docs/curator-widget.md
Health check	docs/health-check.md
Radar and security check	docs/radar.md
Machine memory	docs/machine-memory.md
Backup and extras	docs/backup-and-extras.md
Developer tooling (privacy check, mirror)	docs/developer-tooling.md
Paths, uninstalling, tests	docs/reference.md
Curator internals (German)	curator/README.md
Contributing	CONTRIBUTING.md
Changes (German)	CHANGELOG.md

License

MIT, see LICENSE.