4Runr Project / AI Systems
Active
4Runr OS
Small agents.
One intelligent system.
Your hardware.
A local operating system for coordinated networks of specialized AI agents.
The agent
Think intelligent script.
Each agent has one job. It checks. It decides. It follows a protocol. Then it reports back.
Example — Infrastructure Monitor
Check
Is the service responding?
Decide
Is this temporary or an outage?
Protocol
Retry → Diagnose → Warn → Recover → Escalate
Report
Return the result to the system.
The agent handles the local job.
The system
One agent sees a problem.
The system sees the pattern.
Direction
Network Agent
Unusual traffic
Service Agent
Response time increasing
Security Agent
Repeated authentication failures
Main Agent / Orchestrator
This may be one incident, not three unrelated events.
Agent = local decision
Orchestrator = system-level understanding
Governed intelligence
Intelligence needs boundaries.
Giving AI the ability to act is different from giving it the ability to answer.
Direction
- 01Agent observations
- 02Orchestrator
- 03Proposed decision
- 04Jackel Kernel
- 05Allow / Refuse
- 06Shield
- 07Controlled action
Sentinel
The system doesn't just execute. It observes itself.
↑
Jackel Kernel
Jackel decides what authority the system should allow.
- Governance
- Context
- Permissions
- Policy
- Risk
- Allowed actions
Shield
AI proposes. 4Runr verifies. Then the system acts.
Sentinel
The system doesn't just execute. It observes itself.
The bigger bet
Not one enormous AI doing everything.
Many small agents doing specific things well.
The future of agents may be hundreds of small pieces of intelligence embedded throughout real systems.
One watches infrastructure.
One watches security.
One watches communication.
One watches operations.
Each reports upward. One system understands what they mean together.
Agents do the work
Orchestrator sees the system
Jackel governs decisions
Shield protects execution
Sentinel watches everything
User-controlled hardware
The user's own machine becomes the execution environment. Agents stay close to the systems they are responsible for. Different models can be used depending on the job.
Current state
The platform exists. The product path is still being proven.
Built / Foundation
- CLI
- TUI
- Server architecture
- Docker execution environment
- Foundational security systems
Current objective
- Make one complete agent work
Direction
- One working agent
- Several agents
- Coordinated network of agents
Current objective
Make one complete agent work.
Define a job
Launch the agent locally
Detect something real
Make a decision
Interact with orchestration and governance
Execute an authorized protocol
Report the result
Then we make that path take minutes.