Run
Conversations, streamed responses, scheduled runs, delegation, token limits, and cooperative cancellation.
Read runtime architectureOpen source under Apache 2.0
Praxis Agents OS
The open-source operating system for governed AI agents. Build and run agents inside a shared environment with identity, tools, permissions, approvals, schedules, files, and audit.
Early-stage public buildCore runtime and operational foundations are available. Integrations, knowledge, persistent memory, and artifacts remain roadmap work.Review the roadmap ->
Why the Operating Intelligence Layer exists
The missing layer
An agent needs somewhere to operate
A model can produce an answer. An agent framework can orchestrate a run. Organisational work also needs identity, authority, state, review, and a record of what happened.
ModelsOpenAI, Anthropic, Google, local and specialist services
Agent frameworksReasoning, model calls, tools, and orchestration
Praxis Agents OSIdentity, context, authority, schedules, execution, files, and audit
Private implementationYour agents, workflows, views, rules, integrations, and data
Controlled execution
Context, permission, action, record
The operating path is explicit enough to inspect, test, and change. Tool effects do not disappear inside an opaque agent loop.
01 / InitiateA person, schedule, or delegated agent starts a run inside a workspace.
02 / ResolveThe runtime assembles identity, skills, files, model choice, tools, and authority.
03 / GovernTool effects pass through policy; consequential external writes can pause for approval.
04 / DispatchOne typed registry and audited dispatch path executes the selected tool.
05 / InspectRun state, tool activity, approvals, schedules, and audit records remain visible.
Public evidence
What is in the public build now
Explore the working runtime, governance, tools, files, schedules, and audit foundations available in the repository today.
Govern
Workspace identity, roles, approval controls, audit views, tool effects, and principal-derived run envelopes.
Read governance architectureEquip
A typed tool registry, skills, files, model-provider selection, provider-neutral storage, and multimodal input.
Inspect the sourceExtend
Business-system integrations, knowledge, persistent memory, artifacts, and hardened code execution are sequenced on the public roadmap.
Review the roadmapOpen source and private work
One open foundation. Private operating systems on top.
Building on an open framework does not make your application public. Shared infrastructure can improve upstream while the implementation specific to your organisation remains private.
Upstream open source
Shared foundations improve in public
- Core runtime and platform infrastructure
- Governance, approval, audit, and execution mechanisms
- Generally useful framework improvements
- Reusable integrations with external products and model providers
Private implementation
Your operating capability stays specific to you
- Agents, instructions, workflows, and business rules
- Custom views, dashboards, and user experiences
- Organisation-specific tools and internal connectors
- Configuration, credentials, data, outputs, and operating history
Architectural fit
When configuration stops being enough
Enterprise AI subscriptions are often the fastest route to broad employee productivity and company search. Praxis becomes relevant when the organisation needs customer-controlled code, private interfaces, explicit execution policy, provider flexibility, and a roadmap shaped around its own operations.
Read the product thesis ->Enterprise tools give teams powerful general-purpose assistants. Praxis Agents OS gives an organisation an open foundation for building the private agents, interfaces, workflows, and controls specific to how it operates.
Maintainer-led implementation
Build it yourself, or work with the maintainer
Greg Asquith works with organisations to deploy Praxis Agents OS, connect it to private systems, build agents and operational views, and maintain an ongoing roadmap of custom capabilities.