Download Ora

Functional, not promotional. This page assumes you have already been persuaded by the rest of the site and need operational support. If you arrived here looking for the case for Ora, start with the reliability architecture, the AHI argument, and privacy and sovereignty.

Current download: clone Ora from GitHub. This is a source install, not a packaged app installer yet. Cloning the repository gives you the source tree; the install script verifies your machine, prepares model configuration, and starts the Solo profile.

git clone https://github.com/ora-commons/ora.git ~/ora
cd ~/ora
python3 scripts/install.py --profile solo
./start.sh

What Ora is, operationally

Ora is a sovereign AI system that runs locally on the user's machine. It comes with a framework library, a persistent vault, an adversarial verification pipeline, and a dual-pane workspace that operates text and visual canvas as co-equal modalities. The architecture is model-agnostic — any frontier model, any open-weights model, any future model can serve as the backend. The user provides the model relationship; Ora provides the architecture that makes the model useful for sustained work.

The whole stack is released to the public domain. No patents, no copyrights, no licenses, no permissions required. The user owns the install, owns the vault, owns the frameworks, owns the data.

Platforms

macOS — primary tested path. Apple Silicon gets the best local-model support. Run the commands above, then add local models later with python3 scripts/install.py models if you want them.
Windows native — intended path, pending clean Parallels testing. Use PowerShell: git clone https://github.com/ora-commons/ora.git $env:USERPROFILE\ora, cd $env:USERPROFILE\ora, then py -3 scripts\install.py --profile solo.
Windows WSL — supported as a practical fallback for users comfortable with Ubuntu on Windows. It follows the Linux-style commands and may be easier for Python tooling, with less native browser/file integration.
Linux server — API-only/headless path: clone the repo and run ./scripts/install-server.sh. Linux desktop remains a best-effort community path.

System requirements

You need Git, Python 3.11 or newer, at least 5 GB of free disk for the source install, a browser, and internet access. Local model downloads are optional and can require tens of gigabytes. Ora does not include paid model access; you connect provider accounts or run local open-weights models.

Optional API keys

Ora can start without external keys, and every provider can be added later in Settings → External APIs. The installer now makes the case for the recommended providers and can open official key pages while it runs.

Provider	Why add it	Cost posture
OpenRouter	Broad access to cloud models beyond local and free-model limits.	Free models exist; paid models require credits/payment and vary by use.
Tavily	AI-oriented web search for RAG and research workflows.	Free tier, then usage-based plans.
Artificial Analysis	Independent model intelligence for better model selection.	Free model-benchmark API; commercial data exists separately.
Exa / Brave Search API	Additional search sources with different strengths.	Free credits/tiers, then usage-based.
Direct AI providers	Anthropic, OpenAI, Google, Mistral, DeepSeek, Qwen, and others can bypass OpenRouter's gateway markup for their own models.	Usually usage-based and provider-specific.
Speech and image APIs	Cloud transcription, text-to-speech, image, video, and specialist model workflows.	Usually free tier/trial plus usage-based plans.

OpenRouter is strongly recommended but optional. Free models are rate-limited and sometimes unavailable, so adding payment information or direct provider keys is what turns Ora from "works when free capacity is available" into a more reliable daily tool. FRED is not part of the public install recommendation; it is for specialized economic-data work.

Optional install help

A coding agent is not required. It can help read the install log, explain errors, adjust commands for your machine, and verify the result if something goes sideways.

Compare coding-agent options

Tool	Use case	Link
Claude Code	Strong terminal/file assistant; a one-month subscription can be treated as optional install support and cancelled afterward.	Pricing
OpenAI Codex	Codebase navigation, scripted debugging, and repo-aware assistance.	Codex
GitHub Copilot	Familiar editor-based support in VS Code and GitHub workflows.	Plans
OpenCode	Lightweight terminal coding agent with a low-cost paid plan.	OpenCode
Cursor	AI coding editor for users who want an editor-centered workflow.	Cursor
Continue	Open-source assistant extension using the user's configured provider.	Continue
Cline	VS Code agentic coding extension, often used with the user's own API keys.	Cline
Aider	Terminal pair-programming tool that works with many model providers.	Aider

The first-time experience

The first-time experience is still early. The script should get a technical user to a running source install; packaged installers, polished Windows onboarding, and hybrid/organization deployment are future work. The intent is that a new user can cross from "interested visitor" to "active user" without pretending the current prototype is frictionless.

Honest acknowledgment of the project state

The project is at an early version. Active development is ongoing. There will be rough edges. The methodology is solid; the polish is improving. The author runs Ora as their primary cognitive-automation system; the prototype is real, not a demo. The source install is live; packaged installers are not.

For the foundation

Ora is the system the Ora Knowledge Foundation stewards. The Foundation's role is keeping Ora and the framework library free, defending the public-domain status, and supporting the open-source contributor community across the cognitive tools layer. The Foundation does not gate access to Ora; anyone can fork, modify, redistribute, or build alternatives without permission.