Thin core
The core owns routing, context, session, tooling, and execution orchestration. It does not absorb each platform's business rules.
Built for enterprise systems, not just personal agents.
Personal agents such as OpenClaw are optimized for individual tools and user-owned setup. AtlasClaw separates core orchestration from provider contracts so multi-user deployment, governed execution, and enterprise system boundaries can hold in production.
Core principles
Designed around enterprise boundaries.
AtlasClaw stays useful by keeping boundaries stable instead of swallowing every integration concern into the core.
Rich providers
Each provider packages auth behavior, skills, scripts, references, and normalization for a target system.
Permission inheritance
AtlasClaw passes through real user identity boundaries and lets target platforms keep authorization and auditing where they already belong.
Click to enlarge System flow
From channel entrypoints to the core, then outward through providers to enterprise systems, the execution path remains explicit.
System flow
Close preview From channel entrypoints to the core, then outward through providers to enterprise systems, the execution path remains explicit.
Click to enlarge Core composition
API services, session context, the agent engine, tools, and the provider registry form the reusable substrate.
Core composition
Close preview API services, session context, the agent engine, tools, and the provider registry form the reusable substrate.
Click to enlarge Provider reference flow
The SmartCMP flow shows how requests, approvals, and platform execution can be organized inside a provider model.
Provider reference flow
Close preview The SmartCMP flow shows how requests, approvals, and platform execution can be organized inside a provider model.
Runtime path
From request to governed execution.
- Access channels include web UI, embedded panels, chat platforms, and programmatic webhook calls.
- The AtlasClaw core exposes the API layer, session services, provider registry, and agent engine.
- Execution is handed to providers that speak the target platform's auth and operation model.
- External enterprise systems remain the execution target and source of operational truth.
Design choices
Why this architecture exists.
Why not a personal-agent architecture?
Personal agents such as OpenClaw fit individual productivity well. Enterprise environments need shared deployment, governed access, provider contracts, and system-side audit boundaries, which AtlasClaw makes explicit.
Why provider-qualified skills?
Provider-qualified naming prevents collisions, keeps execution explicit, and makes webhook dispatch safer.
Why separate embedded and standalone modes?
Some teams need an in-product AI module. Others need a multi-user cross-system agent layer. AtlasClaw supports both without changing the core mental model.
How do systems gain LLM brains?
Webhook entrypoints and provider-qualified skills let existing systems call AtlasClaw as an AI execution layer, so traditional products can gain LLM capability without rebuilding their own agent stack.
Next
Integrations Continue into the integration model.
Once the architecture is clear, the next step is understanding how AtlasClaw organizes external systems as providers.