# Praxis Agents OS > Prefer the Markdown URLs below for public text representations. HTML remains canonical. Praxis Agents OS is open-source, self-hosted software that turns what a business knows into work that gets done. The organisation keeps its knowledge base, skills, memories, agents, schedules, data and audit history in its own database on its own infrastructure. AI agents put that knowledge to work across Gmail, Outlook Mail, SharePoint, Google Ads, Meta Ads, Google Analytics, Search Console, BigQuery, Airtable and Notion, with per-tool rules that make sensitive actions wait for a person. Apache 2.0. Public build; the first tagged release (v0.1.0) is pending. ## Core Pages - [Praxis Agents OS](https://www.praxis-agents.ai/index.md): What the product does, what you own, how agents put it to work, your rules, the workspace, and how to run it. Canonical HTML: https://www.praxis-agents.ai - [How It Works](https://www.praxis-agents.ai/how-it-works.md): A seven-step walkthrough from adding your knowledge to keeping the result, plus where Praxis sits and what self-hosting involves. Canonical HTML: https://www.praxis-agents.ai/how-it-works - [Integrations](https://www.praxis-agents.ai/integrations.md): 70 tools across Gmail, Outlook Mail, SharePoint, Google Ads, Meta Ads, Google Analytics, Search Console, BigQuery, Airtable and Notion, their approval policies, and what is planned. Canonical HTML: https://www.praxis-agents.ai/integrations - [Documentation](https://www.praxis-agents.ai/docs.md): Quickstart, architecture notes, contributing guide, and project status. Canonical HTML: https://www.praxis-agents.ai/docs - [Articles](https://www.praxis-agents.ai/articles.md): Product thesis, governed-agent architecture, build-in-public writing, and implementation lessons. Canonical HTML: https://www.praxis-agents.ai/articles - [Services](https://www.praxis-agents.ai/services.md): Initial implementation and optional ongoing development from the project's creator, Greg Asquith. Canonical HTML: https://www.praxis-agents.ai/services - [About](https://www.praxis-agents.ai/about.md): Project origin, Greg Asquith's maintainer role, and the legal company boundary. Canonical HTML: https://www.praxis-agents.ai/about - [Contact](https://www.praxis-agents.ai/contact.md): Implementation enquiries and direct email. Canonical HTML: https://www.praxis-agents.ai/contact ## Articles - [Why an Enterprise AI Subscription Is Not the Same as Owning Your AI Operating Layer](https://www.praxis-agents.ai/articles/enterprise-ai-subscription-vs-operating-layer.md): Claude, ChatGPT, and Gemini enterprise products are increasingly capable. The architectural question is whether configuration is enough, or the organisation needs to own and reshape the operating layer itself. Canonical HTML: https://www.praxis-agents.ai/articles/enterprise-ai-subscription-vs-operating-layer - [Building the Operating Intelligence Layer](https://www.praxis-agents.ai/articles/operating-intelligence-layer.md): Why businesses should own their Operating Intelligence Layer, not outsource the translation of business context into action to external tech companies. Canonical HTML: https://www.praxis-agents.ai/articles/operating-intelligence-layer - [What Is Context Rot? Why AI Agents Lose the Thread](https://www.praxis-agents.ai/articles/context-rot.md): Context rot is what happens when long AI runs become crowded with stale outputs and process noise. Learn how explicit state, prompt budgets, and careful compaction reduce drift. Canonical HTML: https://www.praxis-agents.ai/articles/context-rot - [Why AI Agents Give Inconsistent Results](https://www.praxis-agents.ai/articles/why-ai-agents-inconsistent.md): AI agent inconsistency comes from probabilistic generation, changing context, prompt drift, incomplete information, and moving dependencies. Learn how governed systems reduce the risk. Canonical HTML: https://www.praxis-agents.ai/articles/why-ai-agents-inconsistent - [Looking For OpenClaw For Business?](https://www.praxis-agents.ai/articles/openclaw-for-business.md): OpenClaw is a capable self-hosted personal assistant. This guide explains its trust model and what organisations need to add for shared, governed use. Canonical HTML: https://www.praxis-agents.ai/articles/openclaw-for-business - [AI Agents Are Capacity, Not Seats](https://www.praxis-agents.ai/articles/capacity-not-seats.md): Seat-based software prices individual access. Agent systems are often better evaluated as shared operating capacity, with costs tied to usage, controls, and maintenance. Canonical HTML: https://www.praxis-agents.ai/articles/capacity-not-seats - [What Basic Capacity Costs in London](https://www.praxis-agents.ai/articles/cost-of-capacity.md): A transparent 2026 benchmark for the minimum employer cost of a full-time London Living Wage role, with assumptions and official sources. Canonical HTML: https://www.praxis-agents.ai/articles/cost-of-capacity - [An AI Demo Is Not a Production System](https://www.praxis-agents.ai/articles/beyond-the-ai-prototype.md): An AI-built prototype can validate an idea quickly. Production software also needs security, recovery, observability, data controls, and safe change management. Canonical HTML: https://www.praxis-agents.ai/articles/beyond-the-ai-prototype - [What Does Claude Mean? Models, App, Code, and API](https://www.praxis-agents.ai/articles/claude-explained.md): Claude can refer to Anthropic's models, consumer app, developer tools, or API. Learn which surface fits exploration, coding, and software integration. Canonical HTML: https://www.praxis-agents.ai/articles/claude-explained - [Why Praxis Agents OS Is Not a SaaS Product](https://www.praxis-agents.ai/articles/why-not-saas.md): Praxis Agents OS is Apache-licensed open-source software that organisations can inspect, self-host, and extend, not access sold through a proprietary SaaS subscription. Canonical HTML: https://www.praxis-agents.ai/articles/why-not-saas ## Primary Technical Sources - Repository: https://github.com/Greg-Asquith/praxis-agents-os - Runtime architecture: https://github.com/Greg-Asquith/praxis-agents-os/blob/main/docs/architecture/agent-runtime.md - Governance architecture: https://github.com/Greg-Asquith/praxis-agents-os/blob/main/docs/architecture/governance.md - Agent context architecture: https://github.com/Greg-Asquith/praxis-agents-os/blob/main/docs/architecture/agent-context.md - Prompt-injection threat model: https://github.com/Greg-Asquith/praxis-agents-os/blob/main/docs/architecture/threat-model.md - Roadmap (GitHub): https://github.com/Greg-Asquith/praxis-agents-os/blob/main/docs/plans/000_MASTER_ROADMAP.md ## Boundaries - Inspect the repository and technical documentation for implemented behaviour. - The public build includes ten integrations (70 tools, 31 writes locked to Approval), per-tool approval policies, schedules, a knowledge base, memories, artifacts with share links, Code Mode, provider-native code execution, classifiers, and an audit log. The first tagged release, public cloud deployment guides, self-service password reset, user-developed applications, MCP connections, model failover and run replay are still to come. - Praxis Agents OS is the open-source project. Greg Asquith is its creator and maintainer. Gregory Asquith Limited supplies implementation services.